Tag Archives: algorithms

Google Asked to Remove 3 Billion “Pirate” Search Results

Post Syndicated from Ernesto original https://torrentfreak.com/google-asked-to-remove-3-billion-pirate-search-results-171018/

Copyright holders continue to flood Google with DMCA takedown requests, asking the company to remove “pirate links” from its search results.

In recent years the number of reported URLs has exploded, surging to unprecedented heights.

Since Google first started to report the volume of takedown requests in its Transparency Report, the company has been asked to remove more than three billion allegedly infringing search results.

The frequency at which these URLs are reported has increased over the years and at the moment roughly three million ‘pirate’ URLs are submitted per day.

The URLs are sent in by major rightsholders including members of the BPI, RIAA, and various major Hollywood studios. They target a wide variety of sites, over 1.3 million, but a few dozen ‘repeat offenders’ are causing the most trouble.

File-hosting service 4shared.com currently tops the list of most-targeted domains with 66 million URLs, followed by the now-defunct MP3 download site MP3toys.xyz and Rapidgator.net, with 51 and 28 million URLs respectively.

3 billion URLs

Interestingly, the high volume of takedown notices is used as an argument for and against the DMCA process.

While Google believes that the millions of reported URLs per day are a sign that the DMCA takedown process is working correctly, rightsholders believe the volumes are indicative of an unbeatable game of whack-a-mole.

According to some copyright holders, the takedown efforts do little to seriously combat piracy. Various industry groups have therefore asked governments and lawmakers for broad revisions.

Among other things they want advanced technologies and processes to ensure that infringing content doesn’t reappear elsewhere once it’s removed, a so-called “notice and stay down” approach. In addition, Google has often been asked to demote pirate links in search results.

UK music industry group BPI, who are responsible for more than 10% of all the takedown requests on Google, sees the new milestone as an indicator of how much effort its anti-piracy activities take.

“This 3 billion figure shows how hard the creative sector has to work to police its content online and how much time and resource this takes. The BPI is the world’s largest remover of illegal music links from Google, one third of which are on behalf of independent record labels,” Geoff Taylor, BPI’s Chief Executive, informs TF.

However, there is also some progress to report. Earlier this year BPI announced a voluntary partnership with Google and Bing to demote pirate content faster and more effectively for US visitors.

“We now have a voluntary code of practice in place in the UK, facilitated by Government, that requires Google and Bing to work together with the BPI and other creator organizations to develop lasting solutions to the problem of illegal sites gaining popularity in search listings,” Taylor notes.

According to BPI, both Google and Bing have shown that changes to their algorithms can be effective in demoting the worst pirate sites from the top search results and they hope others will follow suit.

“Other intermediaries should follow this lead and take more responsibility to work with creators to reduce the proliferation of illegal links and disrupt the ability of illegal sites to capture consumers and build black market businesses that take money away from creators.”

Agreement or not, there are still plenty of pirate links in search results, so the BPI is still sending out millions of takedown requests per month.

We asked Google for a comment on the new milestone but at the time of writing, we have yet to hear back. In any event, the issue is bound to remain a hot topic during the months and years to come.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.

My Blogging

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/10/my_blogging.html

Blog regulars will notice that I haven’t been posting as much lately as I have in the past. There are two reasons. One, it feels harder to find things to write about. So often it’s the same stories over and over. I don’t like repeating myself. Two, I am busy writing a book. The title is still: Click Here to Kill Everybody: Peril and Promise in a Hyper-Connected World. The book is a year late, and as a very different table of contents than it had in 2016. I have been writing steadily since mid-August. The book is due to the publisher at the end of March 2018, and will be published in the beginning of September.

This is the current table of contents:

  • Introduction: Everything is Becoming a Computer
  • Part 1: The Trends
    • 1. Capitalism Continues to Drive the Internet
    • 2. Customer/User Control is Next
    • 3. Government Surveillance and Control is Also Increasing
    • 4. Cybercrime is More Profitable Than Ever
    • 5. Cyberwar is the New Normal
    • 6. Algorithms, Automation, and Autonomy Bring New Dangers
    • 7. What We Know About Computer Security
    • 8. Agile is Failing as a Security Paradigm
    • 9. Authentication and Identification are Getting Harder
    • 10. Risks are Becoming Catastrophic
  • Part 2: The Solutions
    • 11. We Need to Regulate the Internet of Things
    • 12. We Need to Defend Critical Infrastructure
    • 13. We Need to Prioritize Defense Over Offence
    • 14. We Need to Make Smarter Decisions About Connecting
    • 15. What’s Likely to Happen, and What We Can Do in Response
    • 16. Where Policy Can Go Wrong
  • Conclusion: Technology and Policy, Together

So that’s what’s been happening.

Predict Billboard Top 10 Hits Using RStudio, H2O and Amazon Athena

Post Syndicated from Gopal Wunnava original https://aws.amazon.com/blogs/big-data/predict-billboard-top-10-hits-using-rstudio-h2o-and-amazon-athena/

Success in the popular music industry is typically measured in terms of the number of Top 10 hits artists have to their credit. The music industry is a highly competitive multi-billion dollar business, and record labels incur various costs in exchange for a percentage of the profits from sales and concert tickets.

Predicting the success of an artist’s release in the popular music industry can be difficult. One release may be extremely popular, resulting in widespread play on TV, radio and social media, while another single may turn out quite unpopular, and therefore unprofitable. Record labels need to be selective in their decision making, and predictive analytics can help them with decision making around the type of songs and artists they need to promote.

In this walkthrough, you leverage H2O.ai, Amazon Athena, and RStudio to make predictions on whether a song might make it to the Top 10 Billboard charts. You explore the GLM, GBM, and deep learning modeling techniques using H2O’s rapid, distributed and easy-to-use open source parallel processing engine. RStudio is a popular IDE, licensed either commercially or under AGPLv3, for working with R. This is ideal if you don’t want to connect to a server via SSH and use code editors such as vi to do analytics. RStudio is available in a desktop version, or a server version that allows you to access R via a web browser. RStudio’s Notebooks feature is used to demonstrate the execution of code and output. In addition, this post showcases how you can leverage Athena for query and interactive analysis during the modeling phase. A working knowledge of statistics and machine learning would be helpful to interpret the analysis being performed in this post.

Walkthrough

Your goal is to predict whether a song will make it to the Top 10 Billboard charts. For this purpose, you will be using multiple modeling techniques―namely GLM, GBM and deep learning―and choose the model that is the best fit.

This solution involves the following steps:

  • Install and configure RStudio with Athena
  • Log in to RStudio
  • Install R packages
  • Connect to Athena
  • Create a dataset
  • Create models

Install and configure RStudio with Athena

Use the following AWS CloudFormation stack to install, configure, and connect RStudio on an Amazon EC2 instance with Athena.

Launching this stack creates all required resources and prerequisites:

  • Amazon EC2 instance with Amazon Linux (minimum size of t2.large is recommended)
  • Provisioning of the EC2 instance in an existing VPC and public subnet
  • Installation of Java 8
  • Assignment of an IAM role to the EC2 instance with the required permissions for accessing Athena and Amazon S3
  • Security group allowing access to the RStudio and SSH ports from the internet (I recommend restricting access to these ports)
  • S3 staging bucket required for Athena (referenced within RStudio as ATHENABUCKET)
  • RStudio username and password
  • Setup logs in Amazon CloudWatch Logs (if needed for additional troubleshooting)
  • Amazon EC2 Systems Manager agent, which makes it easy to manage and patch

All AWS resources are created in the US-East-1 Region. To avoid cross-region data transfer fees, launch the CloudFormation stack in the same region. To check the availability of Athena in other regions, see Region Table.

Log in to RStudio

The instance security group has been automatically configured to allow incoming connections on the RStudio port 8787 from any source internet address. You can edit the security group to restrict source IP access. If you have trouble connecting, ensure that port 8787 isn’t blocked by subnet network ACLS or by your outgoing proxy/firewall.

  1. In the CloudFormation stack, choose Outputs, Value, and then open the RStudio URL. You might need to wait for a few minutes until the instance has been launched.
  2. Log in to RStudio with the and password you provided during setup.

Install R packages

Next, install the required R packages from the RStudio console. You can download the R notebook file containing just the code.

#install pacman – a handy package manager for managing installs
if("pacman" %in% rownames(installed.packages()) == FALSE)
{install.packages("pacman")}  
library(pacman)
p_load(h2o,rJava,RJDBC,awsjavasdk)
h2o.init(nthreads = -1)
##  Connection successful!
## 
## R is connected to the H2O cluster: 
##     H2O cluster uptime:         2 hours 42 minutes 
##     H2O cluster version:        3.10.4.6 
##     H2O cluster version age:    4 months and 4 days !!! 
##     H2O cluster name:           H2O_started_from_R_rstudio_hjx881 
##     H2O cluster total nodes:    1 
##     H2O cluster total memory:   3.30 GB 
##     H2O cluster total cores:    4 
##     H2O cluster allowed cores:  4 
##     H2O cluster healthy:        TRUE 
##     H2O Connection ip:          localhost 
##     H2O Connection port:        54321 
##     H2O Connection proxy:       NA 
##     H2O Internal Security:      FALSE 
##     R Version:                  R version 3.3.3 (2017-03-06)
## Warning in h2o.clusterInfo(): 
## Your H2O cluster version is too old (4 months and 4 days)!
## Please download and install the latest version from http://h2o.ai/download/
#install aws sdk if not present (pre-requisite for using Athena with an IAM role)
if (!aws_sdk_present()) {
  install_aws_sdk()
}

load_sdk()
## NULL

Connect to Athena

Next, establish a connection to Athena from RStudio, using an IAM role associated with your EC2 instance. Use ATHENABUCKET to specify the S3 staging directory.

URL <- 'https://s3.amazonaws.com/athena-downloads/drivers/AthenaJDBC41-1.0.1.jar'
fil <- basename(URL)
#download the file into current working directory
if (!file.exists(fil)) download.file(URL, fil)
#verify that the file has been downloaded successfully
list.files()
## [1] "AthenaJDBC41-1.0.1.jar"
drv <- JDBC(driverClass="com.amazonaws.athena.jdbc.AthenaDriver", fil, identifier.quote="'")

con <- jdbcConnection <- dbConnect(drv, 'jdbc:awsathena://athena.us-east-1.amazonaws.com:443/',
                                   s3_staging_dir=Sys.getenv("ATHENABUCKET"),
                                   aws_credentials_provider_class="com.amazonaws.auth.DefaultAWSCredentialsProviderChain")

Verify the connection. The results returned depend on your specific Athena setup.

con
## <JDBCConnection>
dbListTables(con)
##  [1] "gdelt"               "wikistats"           "elb_logs_raw_native"
##  [4] "twitter"             "twitter2"            "usermovieratings"   
##  [7] "eventcodes"          "events"              "billboard"          
## [10] "billboardtop10"      "elb_logs"            "gdelthist"          
## [13] "gdeltmaster"         "twitter"             "twitter3"

Create a dataset

For this analysis, you use a sample dataset combining information from Billboard and Wikipedia with Echo Nest data in the Million Songs Dataset. Upload this dataset into your own S3 bucket. The table below provides a description of the fields used in this dataset.

Field Description
year Year that song was released
songtitle Title of the song
artistname Name of the song artist
songid Unique identifier for the song
artistid Unique identifier for the song artist
timesignature Variable estimating the time signature of the song
timesignature_confidence Confidence in the estimate for the timesignature
loudness Continuous variable indicating the average amplitude of the audio in decibels
tempo Variable indicating the estimated beats per minute of the song
tempo_confidence Confidence in the estimate for tempo
key Variable with twelve levels indicating the estimated key of the song (C, C#, B)
key_confidence Confidence in the estimate for key
energy Variable that represents the overall acoustic energy of the song, using a mix of features such as loudness
pitch Continuous variable that indicates the pitch of the song
timbre_0_min thru timbre_11_min Variables that indicate the minimum values over all segments for each of the twelve values in the timbre vector
timbre_0_max thru timbre_11_max Variables that indicate the maximum values over all segments for each of the twelve values in the timbre vector
top10 Indicator for whether or not the song made it to the Top 10 of the Billboard charts (1 if it was in the top 10, and 0 if not)

Create an Athena table based on the dataset

In the Athena console, select the default database, sampled, or create a new database.

Run the following create table statement.

create external table if not exists billboard
(
year int,
songtitle string,
artistname string,
songID string,
artistID string,
timesignature int,
timesignature_confidence double,
loudness double,
tempo double,
tempo_confidence double,
key int,
key_confidence double,
energy double,
pitch double,
timbre_0_min double,
timbre_0_max double,
timbre_1_min double,
timbre_1_max double,
timbre_2_min double,
timbre_2_max double,
timbre_3_min double,
timbre_3_max double,
timbre_4_min double,
timbre_4_max double,
timbre_5_min double,
timbre_5_max double,
timbre_6_min double,
timbre_6_max double,
timbre_7_min double,
timbre_7_max double,
timbre_8_min double,
timbre_8_max double,
timbre_9_min double,
timbre_9_max double,
timbre_10_min double,
timbre_10_max double,
timbre_11_min double,
timbre_11_max double,
Top10 int
)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
STORED AS TEXTFILE
LOCATION 's3://aws-bigdata-blog/artifacts/predict-billboard/data'
;

Inspect the table definition for the ‘billboard’ table that you have created. If you chose a database other than sampledb, replace that value with your choice.

dbGetQuery(con, "show create table sampledb.billboard")
##                                      createtab_stmt
## 1       CREATE EXTERNAL TABLE `sampledb.billboard`(
## 2                                       `year` int,
## 3                               `songtitle` string,
## 4                              `artistname` string,
## 5                                  `songid` string,
## 6                                `artistid` string,
## 7                              `timesignature` int,
## 8                `timesignature_confidence` double,
## 9                                `loudness` double,
## 10                                  `tempo` double,
## 11                       `tempo_confidence` double,
## 12                                       `key` int,
## 13                         `key_confidence` double,
## 14                                 `energy` double,
## 15                                  `pitch` double,
## 16                           `timbre_0_min` double,
## 17                           `timbre_0_max` double,
## 18                           `timbre_1_min` double,
## 19                           `timbre_1_max` double,
## 20                           `timbre_2_min` double,
## 21                           `timbre_2_max` double,
## 22                           `timbre_3_min` double,
## 23                           `timbre_3_max` double,
## 24                           `timbre_4_min` double,
## 25                           `timbre_4_max` double,
## 26                           `timbre_5_min` double,
## 27                           `timbre_5_max` double,
## 28                           `timbre_6_min` double,
## 29                           `timbre_6_max` double,
## 30                           `timbre_7_min` double,
## 31                           `timbre_7_max` double,
## 32                           `timbre_8_min` double,
## 33                           `timbre_8_max` double,
## 34                           `timbre_9_min` double,
## 35                           `timbre_9_max` double,
## 36                          `timbre_10_min` double,
## 37                          `timbre_10_max` double,
## 38                          `timbre_11_min` double,
## 39                          `timbre_11_max` double,
## 40                                     `top10` int)
## 41                             ROW FORMAT DELIMITED 
## 42                         FIELDS TERMINATED BY ',' 
## 43                            STORED AS INPUTFORMAT 
## 44       'org.apache.hadoop.mapred.TextInputFormat' 
## 45                                     OUTPUTFORMAT 
## 46  'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
## 47                                        LOCATION
## 48    's3://aws-bigdata-blog/artifacts/predict-billboard/data'
## 49                                  TBLPROPERTIES (
## 50            'transient_lastDdlTime'='1505484133')

Run a sample query

Next, run a sample query to obtain a list of all songs from Janet Jackson that made it to the Billboard Top 10 charts.

dbGetQuery(con, " SELECT songtitle,artistname,top10   FROM sampledb.billboard WHERE lower(artistname) =     'janet jackson' AND top10 = 1")
##                       songtitle    artistname top10
## 1                       Runaway Janet Jackson     1
## 2               Because Of Love Janet Jackson     1
## 3                         Again Janet Jackson     1
## 4                            If Janet Jackson     1
## 5  Love Will Never Do (Without You) Janet Jackson 1
## 6                     Black Cat Janet Jackson     1
## 7               Come Back To Me Janet Jackson     1
## 8                       Alright Janet Jackson     1
## 9                      Escapade Janet Jackson     1
## 10                Rhythm Nation Janet Jackson     1

Determine how many songs in this dataset are specifically from the year 2010.

dbGetQuery(con, " SELECT count(*)   FROM sampledb.billboard WHERE year = 2010")
##   _col0
## 1   373

The sample dataset provides certain song properties of interest that can be analyzed to gauge the impact to the song’s overall popularity. Look at one such property, timesignature, and determine the value that is the most frequent among songs in the database. Timesignature is a measure of the number of beats and the type of note involved.

Running the query directly may result in an error, as shown in the commented lines below. This error is a result of trying to retrieve a large result set over a JDBC connection, which can cause out-of-memory issues at the client level. To address this, reduce the fetch size and run again.

#t<-dbGetQuery(con, " SELECT timesignature FROM sampledb.billboard")
#Note:  Running the preceding query results in the following error: 
#Error in .jcall(rp, "I", "fetch", stride, block): java.sql.SQLException: The requested #fetchSize is more than the allowed value in Athena. Please reduce the fetchSize and try #again. Refer to the Athena documentation for valid fetchSize values.
# Use the dbSendQuery function, reduce the fetch size, and run again
r <- dbSendQuery(con, " SELECT timesignature     FROM sampledb.billboard")
dftimesignature<- fetch(r, n=-1, block=100)
dbClearResult(r)
## [1] TRUE
table(dftimesignature)
## dftimesignature
##    0    1    3    4    5    7 
##   10  143  503 6787  112   19
nrow(dftimesignature)
## [1] 7574

From the results, observe that 6787 songs have a timesignature of 4.

Next, determine the song with the highest tempo.

dbGetQuery(con, " SELECT songtitle,artistname,tempo   FROM sampledb.billboard WHERE tempo = (SELECT max(tempo) FROM sampledb.billboard) ")
##                   songtitle      artistname   tempo
## 1 Wanna Be Startin' Somethin' Michael Jackson 244.307

Create the training dataset

Your model needs to be trained such that it can learn and make accurate predictions. Split the data into training and test datasets, and create the training dataset first.  This dataset contains all observations from the year 2009 and earlier. You may face the same JDBC connection issue pointed out earlier, so this query uses a fetch size.

#BillboardTrain <- dbGetQuery(con, "SELECT * FROM sampledb.billboard WHERE year <= 2009")
#Running the preceding query results in the following error:-
#Error in .verify.JDBC.result(r, "Unable to retrieve JDBC result set for ", : Unable to retrieve #JDBC result set for SELECT * FROM sampledb.billboard WHERE year <= 2009 (Internal error)
#Follow the same approach as before to address this issue.

r <- dbSendQuery(con, "SELECT * FROM sampledb.billboard WHERE year <= 2009")
BillboardTrain <- fetch(r, n=-1, block=100)
dbClearResult(r)
## [1] TRUE
BillboardTrain[1:2,c(1:3,6:10)]
##   year           songtitle artistname timesignature
## 1 2009 The Awkward Goodbye    Athlete             3
## 2 2009        Rubik's Cube    Athlete             3
##   timesignature_confidence loudness   tempo tempo_confidence
## 1                    0.732   -6.320  89.614   0.652
## 2                    0.906   -9.541 117.742   0.542
nrow(BillboardTrain)
## [1] 7201

Create the test dataset

BillboardTest <- dbGetQuery(con, "SELECT * FROM sampledb.billboard where year = 2010")
BillboardTest[1:2,c(1:3,11:15)]
##   year              songtitle        artistname key
## 1 2010 This Is the House That Doubt Built A Day to Remember  11
## 2 2010        Sticks & Bricks A Day to Remember  10
##   key_confidence    energy pitch timbre_0_min
## 1          0.453 0.9666556 0.024        0.002
## 2          0.469 0.9847095 0.025        0.000
nrow(BillboardTest)
## [1] 373

Convert the training and test datasets into H2O dataframes

train.h2o <- as.h2o(BillboardTrain)
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |=================================================================| 100%
test.h2o <- as.h2o(BillboardTest)
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |=================================================================| 100%

Inspect the column names in your H2O dataframes.

colnames(train.h2o)
##  [1] "year"                     "songtitle"               
##  [3] "artistname"               "songid"                  
##  [5] "artistid"                 "timesignature"           
##  [7] "timesignature_confidence" "loudness"                
##  [9] "tempo"                    "tempo_confidence"        
## [11] "key"                      "key_confidence"          
## [13] "energy"                   "pitch"                   
## [15] "timbre_0_min"             "timbre_0_max"            
## [17] "timbre_1_min"             "timbre_1_max"            
## [19] "timbre_2_min"             "timbre_2_max"            
## [21] "timbre_3_min"             "timbre_3_max"            
## [23] "timbre_4_min"             "timbre_4_max"            
## [25] "timbre_5_min"             "timbre_5_max"            
## [27] "timbre_6_min"             "timbre_6_max"            
## [29] "timbre_7_min"             "timbre_7_max"            
## [31] "timbre_8_min"             "timbre_8_max"            
## [33] "timbre_9_min"             "timbre_9_max"            
## [35] "timbre_10_min"            "timbre_10_max"           
## [37] "timbre_11_min"            "timbre_11_max"           
## [39] "top10"

Create models

You need to designate the independent and dependent variables prior to applying your modeling algorithms. Because you’re trying to predict the ‘top10’ field, this would be your dependent variable and everything else would be independent.

Create your first model using GLM. Because GLM works best with numeric data, you create your model by dropping non-numeric variables. You only use the variables in the dataset that describe the numerical attributes of the song in the logistic regression model. You won’t use these variables:  “year”, “songtitle”, “artistname”, “songid”, or “artistid”.

y.dep <- 39
x.indep <- c(6:38)
x.indep
##  [1]  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
## [24] 29 30 31 32 33 34 35 36 37 38

Create Model 1: All numeric variables

Create Model 1 with the training dataset, using GLM as the modeling algorithm and H2O’s built-in h2o.glm function.

modelh1 <- h2o.glm( y = y.dep, x = x.indep, training_frame = train.h2o, family = "binomial")
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |=====                                                            |   8%
  |                                                                       
  |=================================================================| 100%

Measure the performance of Model 1, using H2O’s built-in performance function.

h2o.performance(model=modelh1,newdata=test.h2o)
## H2OBinomialMetrics: glm
## 
## MSE:  0.09924684
## RMSE:  0.3150347
## LogLoss:  0.3220267
## Mean Per-Class Error:  0.2380168
## AUC:  0.8431394
## Gini:  0.6862787
## R^2:  0.254663
## Null Deviance:  326.0801
## Residual Deviance:  240.2319
## AIC:  308.2319
## 
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0   1    Error     Rate
## 0      255  59 0.187898  =59/314
## 1       17  42 0.288136   =17/59
## Totals 272 101 0.203753  =76/373
## 
## Maximum Metrics: Maximum metrics at their respective thresholds
##                         metric threshold    value idx
## 1                       max f1  0.192772 0.525000 100
## 2                       max f2  0.124912 0.650510 155
## 3                 max f0point5  0.416258 0.612903  23
## 4                 max accuracy  0.416258 0.879357  23
## 5                max precision  0.813396 1.000000   0
## 6                   max recall  0.037579 1.000000 282
## 7              max specificity  0.813396 1.000000   0
## 8             max absolute_mcc  0.416258 0.455251  23
## 9   max min_per_class_accuracy  0.161402 0.738854 125
## 10 max mean_per_class_accuracy  0.124912 0.765006 155
## 
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or ` 
h2o.auc(h2o.performance(modelh1,test.h2o)) 
## [1] 0.8431394

The AUC metric provides insight into how well the classifier is able to separate the two classes. In this case, the value of 0.8431394 indicates that the classification is good. (A value of 0.5 indicates a worthless test, while a value of 1.0 indicates a perfect test.)

Next, inspect the coefficients of the variables in the dataset.

dfmodelh1 <- as.data.frame(h2o.varimp(modelh1))
dfmodelh1
##                       names coefficients sign
## 1              timbre_0_max  1.290938663  NEG
## 2                  loudness  1.262941934  POS
## 3                     pitch  0.616995941  NEG
## 4              timbre_1_min  0.422323735  POS
## 5              timbre_6_min  0.349016024  NEG
## 6                    energy  0.348092062  NEG
## 7             timbre_11_min  0.307331997  NEG
## 8              timbre_3_max  0.302225619  NEG
## 9             timbre_11_max  0.243632060  POS
## 10             timbre_4_min  0.224233951  POS
## 11             timbre_4_max  0.204134342  POS
## 12             timbre_5_min  0.199149324  NEG
## 13             timbre_0_min  0.195147119  POS
## 14 timesignature_confidence  0.179973904  POS
## 15         tempo_confidence  0.144242598  POS
## 16            timbre_10_max  0.137644568  POS
## 17             timbre_7_min  0.126995955  NEG
## 18            timbre_10_min  0.123851179  POS
## 19             timbre_7_max  0.100031481  NEG
## 20             timbre_2_min  0.096127636  NEG
## 21           key_confidence  0.083115820  POS
## 22             timbre_6_max  0.073712419  POS
## 23            timesignature  0.067241917  POS
## 24             timbre_8_min  0.061301881  POS
## 25             timbre_8_max  0.060041698  POS
## 26                      key  0.056158445  POS
## 27             timbre_3_min  0.050825116  POS
## 28             timbre_9_max  0.033733561  POS
## 29             timbre_2_max  0.030939072  POS
## 30             timbre_9_min  0.020708113  POS
## 31             timbre_1_max  0.014228818  NEG
## 32                    tempo  0.008199861  POS
## 33             timbre_5_max  0.004837870  POS
## 34                                    NA <NA>

Typically, songs with heavier instrumentation tend to be louder (have higher values in the variable “loudness”) and more energetic (have higher values in the variable “energy”). This knowledge is helpful for interpreting the modeling results.

You can make the following observations from the results:

  • The coefficient estimates for the confidence values associated with the time signature, key, and tempo variables are positive. This suggests that higher confidence leads to a higher predicted probability of a Top 10 hit.
  • The coefficient estimate for loudness is positive, meaning that mainstream listeners prefer louder songs with heavier instrumentation.
  • The coefficient estimate for energy is negative, meaning that mainstream listeners prefer songs that are less energetic, which are those songs with light instrumentation.

These coefficients lead to contradictory conclusions for Model 1. This could be due to multicollinearity issues. Inspect the correlation between the variables “loudness” and “energy” in the training set.

cor(train.h2o$loudness,train.h2o$energy)
## [1] 0.7399067

This number indicates that these two variables are highly correlated, and Model 1 does indeed suffer from multicollinearity. Typically, you associate a value of -1.0 to -0.5 or 1.0 to 0.5 to indicate strong correlation, and a value of 0.1 to 0.1 to indicate weak correlation. To avoid this correlation issue, omit one of these two variables and re-create the models.

You build two variations of the original model:

  • Model 2, in which you keep “energy” and omit “loudness”
  • Model 3, in which you keep “loudness” and omit “energy”

You compare these two models and choose the model with a better fit for this use case.

Create Model 2: Keep energy and omit loudness

colnames(train.h2o)
##  [1] "year"                     "songtitle"               
##  [3] "artistname"               "songid"                  
##  [5] "artistid"                 "timesignature"           
##  [7] "timesignature_confidence" "loudness"                
##  [9] "tempo"                    "tempo_confidence"        
## [11] "key"                      "key_confidence"          
## [13] "energy"                   "pitch"                   
## [15] "timbre_0_min"             "timbre_0_max"            
## [17] "timbre_1_min"             "timbre_1_max"            
## [19] "timbre_2_min"             "timbre_2_max"            
## [21] "timbre_3_min"             "timbre_3_max"            
## [23] "timbre_4_min"             "timbre_4_max"            
## [25] "timbre_5_min"             "timbre_5_max"            
## [27] "timbre_6_min"             "timbre_6_max"            
## [29] "timbre_7_min"             "timbre_7_max"            
## [31] "timbre_8_min"             "timbre_8_max"            
## [33] "timbre_9_min"             "timbre_9_max"            
## [35] "timbre_10_min"            "timbre_10_max"           
## [37] "timbre_11_min"            "timbre_11_max"           
## [39] "top10"
y.dep <- 39
x.indep <- c(6:7,9:38)
x.indep
##  [1]  6  7  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
## [24] 30 31 32 33 34 35 36 37 38
modelh2 <- h2o.glm( y = y.dep, x = x.indep, training_frame = train.h2o, family = "binomial")
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |=======                                                          |  10%
  |                                                                       
  |=================================================================| 100%

Measure the performance of Model 2.

h2o.performance(model=modelh2,newdata=test.h2o)
## H2OBinomialMetrics: glm
## 
## MSE:  0.09922606
## RMSE:  0.3150017
## LogLoss:  0.3228213
## Mean Per-Class Error:  0.2490554
## AUC:  0.8431933
## Gini:  0.6863867
## R^2:  0.2548191
## Null Deviance:  326.0801
## Residual Deviance:  240.8247
## AIC:  306.8247
## 
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      280 34 0.108280  =34/314
## 1       23 36 0.389831   =23/59
## Totals 303 70 0.152815  =57/373
## 
## Maximum Metrics: Maximum metrics at their respective thresholds
##                         metric threshold    value idx
## 1                       max f1  0.254391 0.558140  69
## 2                       max f2  0.113031 0.647208 157
## 3                 max f0point5  0.413999 0.596026  22
## 4                 max accuracy  0.446250 0.876676  18
## 5                max precision  0.811739 1.000000   0
## 6                   max recall  0.037682 1.000000 283
## 7              max specificity  0.811739 1.000000   0
## 8             max absolute_mcc  0.254391 0.469060  69
## 9   max min_per_class_accuracy  0.141051 0.716561 131
## 10 max mean_per_class_accuracy  0.113031 0.761821 157
## 
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
dfmodelh2 <- as.data.frame(h2o.varimp(modelh2))
dfmodelh2
##                       names coefficients sign
## 1                     pitch  0.700331511  NEG
## 2              timbre_1_min  0.510270513  POS
## 3              timbre_0_max  0.402059546  NEG
## 4              timbre_6_min  0.333316236  NEG
## 5             timbre_11_min  0.331647383  NEG
## 6              timbre_3_max  0.252425901  NEG
## 7             timbre_11_max  0.227500308  POS
## 8              timbre_4_max  0.210663865  POS
## 9              timbre_0_min  0.208516163  POS
## 10             timbre_5_min  0.202748055  NEG
## 11             timbre_4_min  0.197246582  POS
## 12            timbre_10_max  0.172729619  POS
## 13         tempo_confidence  0.167523934  POS
## 14 timesignature_confidence  0.167398830  POS
## 15             timbre_7_min  0.142450727  NEG
## 16             timbre_8_max  0.093377516  POS
## 17            timbre_10_min  0.090333426  POS
## 18            timesignature  0.085851625  POS
## 19             timbre_7_max  0.083948442  NEG
## 20           key_confidence  0.079657073  POS
## 21             timbre_6_max  0.076426046  POS
## 22             timbre_2_min  0.071957831  NEG
## 23             timbre_9_max  0.071393189  POS
## 24             timbre_8_min  0.070225578  POS
## 25                      key  0.061394702  POS
## 26             timbre_3_min  0.048384697  POS
## 27             timbre_1_max  0.044721121  NEG
## 28                   energy  0.039698433  POS
## 29             timbre_5_max  0.039469064  POS
## 30             timbre_2_max  0.018461133  POS
## 31                    tempo  0.013279926  POS
## 32             timbre_9_min  0.005282143  NEG
## 33                                    NA <NA>

h2o.auc(h2o.performance(modelh2,test.h2o)) 
## [1] 0.8431933

You can make the following observations:

  • The AUC metric is 0.8431933.
  • Inspecting the coefficient of the variable energy, Model 2 suggests that songs with high energy levels tend to be more popular. This is as per expectation.
  • As H2O orders variables by significance, the variable energy is not significant in this model.

You can conclude that Model 2 is not ideal for this use , as energy is not significant.

CreateModel 3: Keep loudness but omit energy

colnames(train.h2o)
##  [1] "year"                     "songtitle"               
##  [3] "artistname"               "songid"                  
##  [5] "artistid"                 "timesignature"           
##  [7] "timesignature_confidence" "loudness"                
##  [9] "tempo"                    "tempo_confidence"        
## [11] "key"                      "key_confidence"          
## [13] "energy"                   "pitch"                   
## [15] "timbre_0_min"             "timbre_0_max"            
## [17] "timbre_1_min"             "timbre_1_max"            
## [19] "timbre_2_min"             "timbre_2_max"            
## [21] "timbre_3_min"             "timbre_3_max"            
## [23] "timbre_4_min"             "timbre_4_max"            
## [25] "timbre_5_min"             "timbre_5_max"            
## [27] "timbre_6_min"             "timbre_6_max"            
## [29] "timbre_7_min"             "timbre_7_max"            
## [31] "timbre_8_min"             "timbre_8_max"            
## [33] "timbre_9_min"             "timbre_9_max"            
## [35] "timbre_10_min"            "timbre_10_max"           
## [37] "timbre_11_min"            "timbre_11_max"           
## [39] "top10"
y.dep <- 39
x.indep <- c(6:12,14:38)
x.indep
##  [1]  6  7  8  9 10 11 12 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
## [24] 30 31 32 33 34 35 36 37 38
modelh3 <- h2o.glm( y = y.dep, x = x.indep, training_frame = train.h2o, family = "binomial")
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |========                                                         |  12%
  |                                                                       
  |=================================================================| 100%
perfh3<-h2o.performance(model=modelh3,newdata=test.h2o)
perfh3
## H2OBinomialMetrics: glm
## 
## MSE:  0.0978859
## RMSE:  0.3128672
## LogLoss:  0.3178367
## Mean Per-Class Error:  0.264925
## AUC:  0.8492389
## Gini:  0.6984778
## R^2:  0.2648836
## Null Deviance:  326.0801
## Residual Deviance:  237.1062
## AIC:  303.1062
## 
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      286 28 0.089172  =28/314
## 1       26 33 0.440678   =26/59
## Totals 312 61 0.144772  =54/373
## 
## Maximum Metrics: Maximum metrics at their respective thresholds
##                         metric threshold    value idx
## 1                       max f1  0.273799 0.550000  60
## 2                       max f2  0.125503 0.663265 155
## 3                 max f0point5  0.435479 0.628931  24
## 4                 max accuracy  0.435479 0.882038  24
## 5                max precision  0.821606 1.000000   0
## 6                   max recall  0.038328 1.000000 280
## 7              max specificity  0.821606 1.000000   0
## 8             max absolute_mcc  0.435479 0.471426  24
## 9   max min_per_class_accuracy  0.173693 0.745763 120
## 10 max mean_per_class_accuracy  0.125503 0.775073 155
## 
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
dfmodelh3 <- as.data.frame(h2o.varimp(modelh3))
dfmodelh3
##                       names coefficients sign
## 1              timbre_0_max 1.216621e+00  NEG
## 2                  loudness 9.780973e-01  POS
## 3                     pitch 7.249788e-01  NEG
## 4              timbre_1_min 3.891197e-01  POS
## 5              timbre_6_min 3.689193e-01  NEG
## 6             timbre_11_min 3.086673e-01  NEG
## 7              timbre_3_max 3.025593e-01  NEG
## 8             timbre_11_max 2.459081e-01  POS
## 9              timbre_4_min 2.379749e-01  POS
## 10             timbre_4_max 2.157627e-01  POS
## 11             timbre_0_min 1.859531e-01  POS
## 12             timbre_5_min 1.846128e-01  NEG
## 13 timesignature_confidence 1.729658e-01  POS
## 14             timbre_7_min 1.431871e-01  NEG
## 15            timbre_10_max 1.366703e-01  POS
## 16            timbre_10_min 1.215954e-01  POS
## 17         tempo_confidence 1.183698e-01  POS
## 18             timbre_2_min 1.019149e-01  NEG
## 19           key_confidence 9.109701e-02  POS
## 20             timbre_7_max 8.987908e-02  NEG
## 21             timbre_6_max 6.935132e-02  POS
## 22             timbre_8_max 6.878241e-02  POS
## 23            timesignature 6.120105e-02  POS
## 24                      key 5.814805e-02  POS
## 25             timbre_8_min 5.759228e-02  POS
## 26             timbre_1_max 2.930285e-02  NEG
## 27             timbre_9_max 2.843755e-02  POS
## 28             timbre_3_min 2.380245e-02  POS
## 29             timbre_2_max 1.917035e-02  POS
## 30             timbre_5_max 1.715813e-02  POS
## 31                    tempo 1.364418e-02  NEG
## 32             timbre_9_min 8.463143e-05  NEG
## 33                                    NA <NA>
h2o.sensitivity(perfh3,0.5)
## Warning in h2o.find_row_by_threshold(object, t): Could not find exact
## threshold: 0.5 for this set of metrics; using closest threshold found:
## 0.501855569251422. Run `h2o.predict` and apply your desired threshold on a
## probability column.
## [[1]]
## [1] 0.2033898
h2o.auc(perfh3)
## [1] 0.8492389

You can make the following observations:

  • The AUC metric is 0.8492389.
  • From the confusion matrix, the model correctly predicts that 33 songs will be top 10 hits (true positives). However, it has 26 false positives (songs that the model predicted would be Top 10 hits, but ended up not being Top 10 hits).
  • Loudness has a positive coefficient estimate, meaning that this model predicts that songs with heavier instrumentation tend to be more popular. This is the same conclusion from Model 2.
  • Loudness is significant in this model.

Overall, Model 3 predicts a higher number of top 10 hits with an accuracy rate that is acceptable. To choose the best fit for production runs, record labels should consider the following factors:

  • Desired model accuracy at a given threshold
  • Number of correct predictions for top10 hits
  • Tolerable number of false positives or false negatives

Next, make predictions using Model 3 on the test dataset.

predict.regh <- h2o.predict(modelh3, test.h2o)
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |=================================================================| 100%
print(predict.regh)
##   predict        p0          p1
## 1       0 0.9654739 0.034526052
## 2       0 0.9654748 0.034525236
## 3       0 0.9635547 0.036445318
## 4       0 0.9343579 0.065642149
## 5       0 0.9978334 0.002166601
## 6       0 0.9779949 0.022005078
## 
## [373 rows x 3 columns]
predict.regh$predict
##   predict
## 1       0
## 2       0
## 3       0
## 4       0
## 5       0
## 6       0
## 
## [373 rows x 1 column]
dpr<-as.data.frame(predict.regh)
#Rename the predicted column 
colnames(dpr)[colnames(dpr) == 'predict'] <- 'predict_top10'
table(dpr$predict_top10)
## 
##   0   1 
## 312  61

The first set of output results specifies the probabilities associated with each predicted observation.  For example, observation 1 is 96.54739% likely to not be a Top 10 hit, and 3.4526052% likely to be a Top 10 hit (predict=1 indicates Top 10 hit and predict=0 indicates not a Top 10 hit).  The second set of results list the actual predictions made.  From the third set of results, this model predicts that 61 songs will be top 10 hits.

Compute the baseline accuracy, by assuming that the baseline predicts the most frequent outcome, which is that most songs are not Top 10 hits.

table(BillboardTest$top10)
## 
##   0   1 
## 314  59

Now observe that the baseline model would get 314 observations correct, and 59 wrong, for an accuracy of 314/(314+59) = 0.8418231.

It seems that Model 3, with an accuracy of 0.8552, provides you with a small improvement over the baseline model. But is this model useful for record labels?

View the two models from an investment perspective:

  • A production company is interested in investing in songs that are more likely to make it to the Top 10. The company’s objective is to minimize the risk of financial losses attributed to investing in songs that end up unpopular.
  • How many songs does Model 3 correctly predict as a Top 10 hit in 2010? Looking at the confusion matrix, you see that it predicts 33 top 10 hits correctly at an optimal threshold, which is more than half the number
  • It will be more useful to the record label if you can provide the production company with a list of songs that are highly likely to end up in the Top 10.
  • The baseline model is not useful, as it simply does not label any song as a hit.

Considering the three models built so far, you can conclude that Model 3 proves to be the best investment choice for the record label.

GBM model

H2O provides you with the ability to explore other learning models, such as GBM and deep learning. Explore building a model using the GBM technique, using the built-in h2o.gbm function.

Before you do this, you need to convert the target variable to a factor for multinomial classification techniques.

train.h2o$top10=as.factor(train.h2o$top10)
gbm.modelh <- h2o.gbm(y=y.dep, x=x.indep, training_frame = train.h2o, ntrees = 500, max_depth = 4, learn_rate = 0.01, seed = 1122,distribution="multinomial")
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |===                                                              |   5%
  |                                                                       
  |=====                                                            |   7%
  |                                                                       
  |======                                                           |   9%
  |                                                                       
  |=======                                                          |  10%
  |                                                                       
  |======================                                           |  33%
  |                                                                       
  |=====================================                            |  56%
  |                                                                       
  |====================================================             |  79%
  |                                                                       
  |================================================================ |  98%
  |                                                                       
  |=================================================================| 100%
perf.gbmh<-h2o.performance(gbm.modelh,test.h2o)
perf.gbmh
## H2OBinomialMetrics: gbm
## 
## MSE:  0.09860778
## RMSE:  0.3140188
## LogLoss:  0.3206876
## Mean Per-Class Error:  0.2120263
## AUC:  0.8630573
## Gini:  0.7261146
## 
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      266 48 0.152866  =48/314
## 1       16 43 0.271186   =16/59
## Totals 282 91 0.171582  =64/373
## 
## Maximum Metrics: Maximum metrics at their respective thresholds
##                       metric threshold    value idx
## 1                     max f1  0.189757 0.573333  90
## 2                     max f2  0.130895 0.693717 145
## 3               max f0point5  0.327346 0.598802  26
## 4               max accuracy  0.442757 0.876676  14
## 5              max precision  0.802184 1.000000   0
## 6                 max recall  0.049990 1.000000 284
## 7            max specificity  0.802184 1.000000   0
## 8           max absolute_mcc  0.169135 0.496486 104
## 9 max min_per_class_accuracy  0.169135 0.796610 104
## 10 max mean_per_class_accuracy  0.169135 0.805948 104
## 
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `
h2o.sensitivity(perf.gbmh,0.5)
## Warning in h2o.find_row_by_threshold(object, t): Could not find exact
## threshold: 0.5 for this set of metrics; using closest threshold found:
## 0.501205344484314. Run `h2o.predict` and apply your desired threshold on a
## probability column.
## [[1]]
## [1] 0.1355932
h2o.auc(perf.gbmh)
## [1] 0.8630573

This model correctly predicts 43 top 10 hits, which is 10 more than the number predicted by Model 3. Moreover, the AUC metric is higher than the one obtained from Model 3.

As seen above, H2O’s API provides the ability to obtain key statistical measures required to analyze the models easily, using several built-in functions. The record label can experiment with different parameters to arrive at the model that predicts the maximum number of Top 10 hits at the desired level of accuracy and threshold.

H2O also allows you to experiment with deep learning models. Deep learning models have the ability to learn features implicitly, but can be more expensive computationally.

Now, create a deep learning model with the h2o.deeplearning function, using the same training and test datasets created before. The time taken to run this model depends on the type of EC2 instance chosen for this purpose.  For models that require more computation, consider using accelerated computing instances such as the P2 instance type.

system.time(
  dlearning.modelh <- h2o.deeplearning(y = y.dep,
                                      x = x.indep,
                                      training_frame = train.h2o,
                                      epoch = 250,
                                      hidden = c(250,250),
                                      activation = "Rectifier",
                                      seed = 1122,
                                      distribution="multinomial"
  )
)
## 
  |                                                                       
  |                                                                 |   0%
  |                                                                       
  |===                                                              |   4%
  |                                                                       
  |=====                                                            |   8%
  |                                                                       
  |========                                                         |  12%
  |                                                                       
  |==========                                                       |  16%
  |                                                                       
  |=============                                                    |  20%
  |                                                                       
  |================                                                 |  24%
  |                                                                       
  |==================                                               |  28%
  |                                                                       
  |=====================                                            |  32%
  |                                                                       
  |=======================                                          |  36%
  |                                                                       
  |==========================                                       |  40%
  |                                                                       
  |=============================                                    |  44%
  |                                                                       
  |===============================                                  |  48%
  |                                                                       
  |==================================                               |  52%
  |                                                                       
  |====================================                             |  56%
  |                                                                       
  |=======================================                          |  60%
  |                                                                       
  |==========================================                       |  64%
  |                                                                       
  |============================================                     |  68%
  |                                                                       
  |===============================================                  |  72%
  |                                                                       
  |=================================================                |  76%
  |                                                                       
  |====================================================             |  80%
  |                                                                       
  |=======================================================          |  84%
  |                                                                       
  |=========================================================        |  88%
  |                                                                       
  |============================================================     |  92%
  |                                                                       
  |==============================================================   |  96%
  |                                                                       
  |=================================================================| 100%
##    user  system elapsed 
##   1.216   0.020 166.508
perf.dl<-h2o.performance(model=dlearning.modelh,newdata=test.h2o)
perf.dl
## H2OBinomialMetrics: deeplearning
## 
## MSE:  0.1678359
## RMSE:  0.4096778
## LogLoss:  1.86509
## Mean Per-Class Error:  0.3433013
## AUC:  0.7568822
## Gini:  0.5137644
## 
## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
##          0  1    Error     Rate
## 0      290 24 0.076433  =24/314
## 1       36 23 0.610169   =36/59
## Totals 326 47 0.160858  =60/373
## 
## Maximum Metrics: Maximum metrics at their respective thresholds
##                       metric threshold    value idx
## 1                     max f1  0.826267 0.433962  46
## 2                     max f2  0.000000 0.588235 239
## 3               max f0point5  0.999929 0.511811  16
## 4               max accuracy  0.999999 0.865952  10
## 5              max precision  1.000000 1.000000   0
## 6                 max recall  0.000000 1.000000 326
## 7            max specificity  1.000000 1.000000   0
## 8           max absolute_mcc  0.999929 0.363219  16
## 9 max min_per_class_accuracy  0.000004 0.662420 145
## 10 max mean_per_class_accuracy  0.000000 0.685334 224
## 
## Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
h2o.sensitivity(perf.dl,0.5)
## Warning in h2o.find_row_by_threshold(object, t): Could not find exact
## threshold: 0.5 for this set of metrics; using closest threshold found:
## 0.496293348880151. Run `h2o.predict` and apply your desired threshold on a
## probability column.
## [[1]]
## [1] 0.3898305
h2o.auc(perf.dl)
## [1] 0.7568822

The AUC metric for this model is 0.7568822, which is less than what you got from the earlier models. I recommend further experimentation using different hyper parameters, such as the learning rate, epoch or the number of hidden layers.

H2O’s built-in functions provide many key statistical measures that can help measure model performance. Here are some of these key terms.

Metric Description
Sensitivity Measures the proportion of positives that have been correctly identified. It is also called the true positive rate, or recall.
Specificity Measures the proportion of negatives that have been correctly identified. It is also called the true negative rate.
Threshold Cutoff point that maximizes specificity and sensitivity. While the model may not provide the highest prediction at this point, it would not be biased towards positives or negatives.
Precision The fraction of the documents retrieved that are relevant to the information needed, for example, how many of the positively classified are relevant
AUC

Provides insight into how well the classifier is able to separate the two classes. The implicit goal is to deal with situations where the sample distribution is highly skewed, with a tendency to overfit to a single class.

0.90 – 1 = excellent (A)

0.8 – 0.9 = good (B)

0.7 – 0.8 = fair (C)

.6 – 0.7 = poor (D)

0.5 – 0.5 = fail (F)

Here’s a summary of the metrics generated from H2O’s built-in functions for the three models that produced useful results.

Metric Model 3 GBM Model Deep Learning Model

Accuracy

(max)

0.882038

(t=0.435479)

0.876676

(t=0.442757)

0.865952

(t=0.999999)

Precision

(max)

1.0

(t=0.821606)

1.0

(t=0802184)

1.0

(t=1.0)

Recall

(max)

1.0 1.0

1.0

(t=0)

Specificity

(max)

1.0 1.0

1.0

(t=1)

Sensitivity

 

0.2033898 0.1355932

0.3898305

(t=0.5)

AUC 0.8492389 0.8630573 0.756882

Note: ‘t’ denotes threshold.

Your options at this point could be narrowed down to Model 3 and the GBM model, based on the AUC and accuracy metrics observed earlier.  If the slightly lower accuracy of the GBM model is deemed acceptable, the record label can choose to go to production with the GBM model, as it can predict a higher number of Top 10 hits.  The AUC metric for the GBM model is also higher than that of Model 3.

Record labels can experiment with different learning techniques and parameters before arriving at a model that proves to be the best fit for their business. Because deep learning models can be computationally expensive, record labels can choose more powerful EC2 instances on AWS to run their experiments faster.

Conclusion

In this post, I showed how the popular music industry can use analytics to predict the type of songs that make the Top 10 Billboard charts. By running H2O’s scalable machine learning platform on AWS, data scientists can easily experiment with multiple modeling techniques and interactively query the data using Amazon Athena, without having to manage the underlying infrastructure. This helps record labels make critical decisions on the type of artists and songs to promote in a timely fashion, thereby increasing sales and revenue.

If you have questions or suggestions, please comment below.


Additional Reading

Learn how to build and explore a simple geospita simple GEOINT application using SparkR.


About the Authors

gopalGopal Wunnava is a Partner Solution Architect with the AWS GSI Team. He works with partners and customers on big data engagements, and is passionate about building analytical solutions that drive business capabilities and decision making. In his spare time, he loves all things sports and movies related and is fond of old classics like Asterix, Obelix comics and Hitchcock movies.

 

 

Bob Strahan, a Senior Consultant with AWS Professional Services, contributed to this post.

 

 

Technology to Out Sex Workers

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/10/technology_to_o.html

Two related stories:

PornHub is using machine learning algorithms to identify actors in different videos, so as to better index them. People are worried that it can really identify them, by linking their stage names to their real names.

Facebook somehow managed to link a sex worker’s clients under her fake name to her real profile.

Sometimes people have legitimate reasons for having two identities. That is becoming harder and harder.

Introducing Gluon: a new library for machine learning from AWS and Microsoft

Post Syndicated from Ana Visneski original https://aws.amazon.com/blogs/aws/introducing-gluon-a-new-library-for-machine-learning-from-aws-and-microsoft/

Post by Dr. Matt Wood

Today, AWS and Microsoft announced Gluon, a new open source deep learning interface which allows developers to more easily and quickly build machine learning models, without compromising performance.

Gluon Logo

Gluon provides a clear, concise API for defining machine learning models using a collection of pre-built, optimized neural network components. Developers who are new to machine learning will find this interface more familiar to traditional code, since machine learning models can be defined and manipulated just like any other data structure. More seasoned data scientists and researchers will value the ability to build prototypes quickly and utilize dynamic neural network graphs for entirely new model architectures, all without sacrificing training speed.

Gluon is available in Apache MXNet today, a forthcoming Microsoft Cognitive Toolkit release, and in more frameworks over time.

Neural Networks vs Developers
Machine learning with neural networks (including ‘deep learning’) has three main components: data for training; a neural network model, and an algorithm which trains the neural network. You can think of the neural network in a similar way to a directed graph; it has a series of inputs (which represent the data), which connect to a series of outputs (the prediction), through a series of connected layers and weights. During training, the algorithm adjusts the weights in the network based on the error in the network output. This is the process by which the network learns; it is a memory and compute intensive process which can take days.

Deep learning frameworks such as Caffe2, Cognitive Toolkit, TensorFlow, and Apache MXNet are, in part, an answer to the question ‘how can we speed this process up? Just like query optimizers in databases, the more a training engine knows about the network and the algorithm, the more optimizations it can make to the training process (for example, it can infer what needs to be re-computed on the graph based on what else has changed, and skip the unaffected weights to speed things up). These frameworks also provide parallelization to distribute the computation process, and reduce the overall training time.

However, in order to achieve these optimizations, most frameworks require the developer to do some extra work: specifically, by providing a formal definition of the network graph, up-front, and then ‘freezing’ the graph, and just adjusting the weights.

The network definition, which can be large and complex with millions of connections, usually has to be constructed by hand. Not only are deep learning networks unwieldy, but they can be difficult to debug and it’s hard to re-use the code between projects.

The result of this complexity can be difficult for beginners and is a time-consuming task for more experienced researchers. At AWS, we’ve been experimenting with some ideas in MXNet around new, flexible, more approachable ways to define and train neural networks. Microsoft is also a contributor to the open source MXNet project, and were interested in some of these same ideas. Based on this, we got talking, and found we had a similar vision: to use these techniques to reduce the complexity of machine learning, making it accessible to more developers.

Enter Gluon: dynamic graphs, rapid iteration, scalable training
Gluon introduces four key innovations.

  1. Friendly API: Gluon networks can be defined using a simple, clear, concise code – this is easier for developers to learn, and much easier to understand than some of the more arcane and formal ways of defining networks and their associated weighted scoring functions.
  2. Dynamic networks: the network definition in Gluon is dynamic: it can bend and flex just like any other data structure. This is in contrast to the more common, formal, symbolic definition of a network which the deep learning framework has to effectively carve into stone in order to be able to effectively optimizing computation during training. Dynamic networks are easier to manage, and with Gluon, developers can easily ‘hybridize’ between these fast symbolic representations and the more friendly, dynamic ‘imperative’ definitions of the network and algorithms.
  3. The algorithm can define the network: the model and the training algorithm are brought much closer together. Instead of separate definitions, the algorithm can adjust the network dynamically during definition and training. Not only does this mean that developers can use standard programming loops, and conditionals to create these networks, but researchers can now define even more sophisticated algorithms and models which were not possible before. They are all easier to create, change, and debug.
  4. High performance operators for training: which makes it possible to have a friendly, concise API and dynamic graphs, without sacrificing training speed. This is a huge step forward in machine learning. Some frameworks bring a friendly API or dynamic graphs to deep learning, but these previous methods all incur a cost in terms of training speed. As with other areas of software, abstraction can slow down computation since it needs to be negotiated and interpreted at run time. Gluon can efficiently blend together a concise API with the formal definition under the hood, without the developer having to know about the specific details or to accommodate the compiler optimizations manually.

The team here at AWS, and our collaborators at Microsoft, couldn’t be more excited to bring these improvements to developers through Gluon. We’re already seeing quite a bit of excitement from developers and researchers alike.

Getting started with Gluon
Gluon is available today in Apache MXNet, with support coming for the Microsoft Cognitive Toolkit in a future release. We’re also publishing the front-end interface and the low-level API specifications so it can be included in other frameworks in the fullness of time.

You can get started with Gluon today. Fire up the AWS Deep Learning AMI with a single click and jump into one of 50 fully worked, notebook examples. If you’re a contributor to a machine learning framework, check out the interface specs on GitHub.

-Dr. Matt Wood

Backing Up WordPress

Post Syndicated from Roderick Bauer original https://www.backblaze.com/blog/backing-up-wordpress/

WordPress cloud backup
WordPress logo

WordPress is the most popular CMS (Content Management System) for websites, with almost 30% of all websites in the world using WordPress. That’s a lot of sites — over 350 million!

In this post we’ll talk about the different approaches to keeping the data on your WordPress website safe.


Stop the Presses! (Or the Internet!)

As we were getting ready to publish this post, we received news from UpdraftPlus, one of the biggest WordPress plugin developers, that they are supporting Backblaze B2 as a storage solution for their backup plugin. They shipped the update (1.13.9) this week. This is great news for Backblaze customers! UpdraftPlus is also offering a 20% discount to Backblaze customers wishing to purchase or upgrade to UpdraftPlus Premium. The complete information is below.

UpdraftPlus joins backup plugin developer XCloner — Backup and Restore in supporting Backblaze B2. A third developer, BlogVault, also announced their intent to support Backblaze B2. Contact your favorite WordPress backup plugin developer and urge them to support Backblaze B2, as well.

Now, back to our post…


Your WordPress website data is on a web server that’s most likely located in a large data center. You might wonder why it is necessary to have a backup of your website if it’s in a data center. Website data can be lost in a number of ways, including mistakes by the website owner (been there), hacking, or even domain ownership dispute (I’ve seen it happen more than once). A website backup also can provide a history of changes you’ve made to the website, which can be useful. As an overall strategy, it’s best to have a backup of any data that you can’t afford to lose for personal or business reasons.

Your web hosting company might provide backup services as part of your hosting plan. If you are using their service, you should know where and how often your data is being backed up. You don’t want to find out too late that your backup plan was not adequate.

Sites on WordPress.com are automatically backed up by VaultPress (Automattic), which also is available for self-hosted WordPress installations. If you don’t want the work or decisions involved in managing the hosting for your WordPress site, WordPress.com will handle it for you. You do, however, give up some customization abilities, such as the option to add plugins of your own choice.

Very large and active websites might consider WordPress VIP by Automattic, or another premium WordPress hosting service such as Pagely.com.

This post is about backing up self-hosted WordPress sites, so we’ll focus on those options.

WordPress Backup

Backup strategies for WordPress can be divided into broad categories depending on 1) what you back up, 2) when you back up, and 3) where the data is backed up.

With server data, such as with a WordPress installation, you should plan to have three copies of the data (the 3-2-1 backup strategy). The first is the active data on the WordPress web server, the second is a backup stored on the web server or downloaded to your local computer, and the third should be in another location, such as the cloud.

We’ll talk about the different approaches to backing up WordPress, but we recommend using a WordPress plugin to handle your backups. A backup plugin can automate the task, optimize your backup storage space, and alert you of problems with your backups or WordPress itself. We’ll cover plugins in more detail, below.

What to Back Up?

The main components of your WordPress installation are:

You should decide which of these elements you wish to back up. The database is the top priority, as it contains all your website posts and pages (exclusive of media). Your current theme is important, as it likely contains customizations you’ve made. Following those in priority are any other files you’ve customized or made changes to.

You can choose to back up the WordPress core installation and plugins, if you wish, but these files can be downloaded again if necessary from the source, so you might not wish to include them. You likely have all the media files you use on your website on your local computer (which should be backed up), so it is your choice whether to back these up from the server as well.

If you wish to be able to recreate your entire website easily in case of data loss or disaster, you might choose to back up everything, though on a large website this could be a lot of data.

Generally, you should 1) prioritize any file that you’ve customized that you can’t afford to lose, and 2) decide whether you need a copy of everything in order to get your site back up quickly. These choices will determine your backup method and the amount of storage you need.

A good backup plugin for WordPress enables you to specify which files you wish to back up, and even to create separate backups and schedules for different backup contents. That’s another good reason to use a plugin for backing up WordPress.

When to Back Up?

You can back up manually at any time by using the Export tool in WordPress. This is handy if you wish to do a quick backup of your site or parts of it. Since it is manual, however, it is not a part of a dependable backup plan that should be done regularly. If you wish to use this tool, go to Tools, Export, and select what you wish to back up. The output will be an XML file that uses the WordPress Extended RSS format, also known as WXR. You can create a WXR file that contains all of the information on your site or just portions of the site, such as posts or pages by selecting: All content, Posts, Pages, or Media.
Note: You can use WordPress’s Export tool for sites hosted on WordPress.com, as well.

Export instruction for WordPress

Many of the backup plugins we’ll be discussing later also let you do a manual backup on demand in addition to regularly scheduled or continuous backups.

Note:  Another use of the WordPress Export tool and the WXR file is to transfer or clone your website to another server. Once you have exported the WXR file from the website you wish to transfer from, you can import the WXR file from the Tools, Import menu on the new WordPress destination site. Be aware that there are file size limits depending on the settings on your web server. See the WordPress Codex entry for more information. To make this job easier, you may wish to use one of a number of WordPress plugins designed specifically for this task.

You also can manually back up the WordPress MySQL database using a number of tools or a plugin. The WordPress Codex has good information on this. All WordPress plugins will handle this for you and do it automatically. They also typically include tools for optimizing the database tables, which is just good housekeeping.

A dependable backup strategy doesn’t rely on manual backups, which means you should consider using one of the many backup plugins available either free or for purchase. We’ll talk more about them below.

Which Format To Back Up In?

In addition to the WordPress WXR format, plugins and server tools will use various file formats and compression algorithms to store and compress your backup. You may get to choose between zip, tar, tar.gz, tar.gz2, and others. See The Most Common Archive File Formats for more information on these formats.

Select a format that you know you can access and unarchive should you need access to your backup. All of these formats are standard and supported across operating systems, though you might need to download a utility to access the file.

Where To Back Up?

Once you have your data in a suitable format for backup, where do you back it up to?

We want to have multiple copies of our active website data, so we’ll choose more than one destination for our backup data. The backup plugins we’ll discuss below enable you to specify one or more possible destinations for your backup. The possible destinations for your backup include:

A backup folder on your web server
A backup folder on your web server is an OK solution if you also have a copy elsewhere. Depending on your hosting plan, the size of your site, and what you include in the backup, you may or may not have sufficient disk space on the web server. Some backup plugins allow you to configure the plugin to keep only a certain number of recent backups and delete older ones, saving you disk space on the server.
Email to you
Because email servers have size limitations, the email option is not the best one to use unless you use it to specifically back up just the database or your main theme files.
FTP, SFTP, SCP, WebDAV
FTP, SFTP, SCP, and WebDAV are all widely-supported protocols for transferring files over the internet and can be used if you have access credentials to another server or supported storage device that is suitable for storing a backup.
Sync service (Dropbox, SugarSync, Google Drive, OneDrive)
A sync service is another possible server storage location though it can be a pricier choice depending on the plan you have and how much you wish to store.
Cloud storage (Backblaze B2, Amazon S3, Google Cloud, Microsoft Azure, Rackspace)
A cloud storage service can be an inexpensive and flexible option with pay-as-you go pricing for storing backups and other data.

A good website backup strategy would be to have multiple backups of your website data: one in a backup folder on your web hosting server, one downloaded to your local computer, and one in the cloud, such as with Backblaze B2.

If I had to choose just one of these, I would choose backing up to the cloud because it is geographically separated from both your local computer and your web host, it uses fault-tolerant and redundant data storage technologies to protect your data, and it is available from anywhere if you need to restore your site.

Backup Plugins for WordPress

Probably the easiest and most common way to implement a solid backup strategy for WordPress is to use one of the many backup plugins available for WordPress. Fortunately, there are a number of good ones and are available free or in “freemium” plans in which you can use the free version and pay for more features and capabilities only if you need them. The premium options can give you more flexibility in configuring backups or have additional options for where you can store the backups.

How to Choose a WordPress Backup Plugin

screenshot of WordPress plugins search

When considering which plugin to use, you should take into account a number of factors in making your choice.

Is the plugin actively maintained and up-to-date? You can determine this from the listing in the WordPress Plugin Repository. You also can look at reviews and support comments to get an idea of user satisfaction and how well issues are resolved.

Does the plugin work with your web hosting provider? Generally, well-supported plugins do, but you might want to check to make sure there are no issues with your hosting provider.

Does it support the cloud service or protocol you wish to use? This can be determined from looking at the listing in the WordPress Plugin Repository or on the developer’s website. Developers often will add support for cloud services or other backup destinations based on user demand, so let the developer know if there is a feature or backup destination you’d like them to add to their plugin.

Other features and options to consider in choosing a backup plugin are:

  • Whether encryption of your backup data is available
  • What are the options for automatically deleting backups from the storage destination?
  • Can you globally exclude files, folders, and specific types of files from the backup?
  • Do the options for scheduling automatic backups meet your needs for frequency?
  • Can you exclude/include specific database tables (a good way to save space in your backup)?

WordPress Backup Plugins Review

Let’s review a few of the top choices for WordPress backup plugins.

UpdraftPlus

UpdraftPlus is one of the most popular backup plugins for WordPress with over one million active installations. It is available in both free and Premium versions.

UpdraftPlus just released support for Backblaze B2 Cloud Storage in their 1.13.9 update on September 25. According to the developer, support for Backblaze B2 was the most frequent request for a new storage option for their plugin. B2 support is available in their Premium plugin and as a stand-alone update to their standard product.

Note: The developers of UpdraftPlus are offering a special 20% discount to Backblaze customers on the purchase of UpdraftPlus Premium by using the coupon code backblaze20. The discount is valid until the end of Friday, October 6th, 2017.

screenshot of Backblaze B2 cloud backup for WordPress in UpdraftPlus

XCloner — Backup and Restore

XCloner — Backup and Restore is a useful open-source plugin with many options for backing up WordPress.

XCloner supports B2 Cloud Storage in their free plugin.

screenshot of XCloner WordPress Backblaze B2 backup settings

BlogVault

BlogVault describes themselves as a “complete WordPress backup solution.” They offer a free trial of their paid WordPress backup subscription service that features real-time backups of changes to your WordPress site, as well as many other features.

BlogVault has announced their intent to support Backblaze B2 Cloud Storage in a future update.

screenshot of BlogValut WordPress Backup settings

BackWPup

BackWPup is a popular and free option for backing up WordPress. It supports a number of options for storing your backup, including the cloud, FTP, email, or on your local computer.

screenshot of BackWPup WordPress backup settings

WPBackItUp

WPBackItUp has been around since 2012 and is highly rated. It has both free and paid versions.

screenshot of WPBackItUp WordPress backup settings

VaultPress

VaultPress is part of Automattic’s well-known WordPress product, JetPack. You will need a JetPack subscription plan to use VaultPress. There are different pricing plans with different sets of features.

screenshot of VaultPress backup settings

Backup by Supsystic

Backup by Supsystic supports a number of options for backup destinations, encryption, and scheduling.

screenshot of Backup by Supsystic backup settings

BackupWordPress

BackUpWordPress is an open-source project on Github that has a popular and active following and many positive reviews.

screenshot of BackupWordPress WordPress backup settings

BackupBuddy

BackupBuddy, from iThemes, is the old-timer of backup plugins, having been around since 2010. iThemes knows a lot about WordPress, as they develop plugins, themes, utilities, and provide training in WordPress.

BackupBuddy’s backup includes all WordPress files, all files in the WordPress Media library, WordPress themes, and plugins. BackupBuddy generates a downloadable zip file of the entire WordPress website. Remote storage destinations also are supported.

screenshot of BackupBuddy settings

WordPress and the Cloud

Do you use WordPress and back up to the cloud? We’d like to hear about it. We’d also like to hear whether you are interested in using B2 Cloud Storage for storing media files served by WordPress. If you are, we’ll write about it in a future post.

In the meantime, keep your eye out for new plugins supporting Backblaze B2, or better yet, urge them to support B2 if they’re not already.

The Best Backup Strategy is the One You Use

There are other approaches and tools for backing up WordPress that you might use. If you have an approach that works for you, we’d love to hear about it in the comments.

The post Backing Up WordPress appeared first on Backblaze Blog | Cloud Storage & Cloud Backup.

ISO Rejects NSA Encryption Algorithms

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/09/iso_rejects_nsa.html

The ISO has decided not to approve two NSA-designed block encryption algorithms: Speck and Simon. It’s because the NSA is not trusted to put security ahead of surveillance:

A number of them voiced their distrust in emails to one another, seen by Reuters, and in written comments that are part of the process. The suspicions stem largely from internal NSA documents disclosed by Snowden that showed the agency had previously plotted to manipulate standards and promote technology it could penetrate. Budget documents, for example, sought funding to “insert vulnerabilities into commercial encryption systems.”

More than a dozen of the experts involved in the approval process for Simon and Speck feared that if the NSA was able to crack the encryption techniques, it would gain a “back door” into coded transmissions, according to the interviews and emails and other documents seen by Reuters.

“I don’t trust the designers,” Israeli delegate Orr Dunkelman, a computer science professor at the University of Haifa, told Reuters, citing Snowden’s papers. “There are quite a lot of people in NSA who think their job is to subvert standards. My job is to secure standards.”

I don’t trust the NSA, either.

Confusing Self-Driving Cars by Altering Road Signs

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/08/confusing_self-.html

Researchers found that they could confuse the road sign detection algorithms of self-driving cars by adding stickers to the signs on the road. They could, for example, cause a car to think that a stop sign is a 45 mph speed limit sign. The changes are subtle, though — look at the photo from the article.

Research paper:

Robust Physical-World Attacks on Machine Learning Models,” by Ivan Evtimov, Kevin Eykholt, Earlence Fernandes, Tadayoshi Kohno, Bo Li, Atul Prakash, Amir Rahmati, and Dawn Song:

Abstract: Deep neural network-based classifiers are known to be vulnerable to adversarial examples that can fool them into misclassifying their input through the addition of small-magnitude perturbations. However, recent studies have demonstrated that such adversarial examples are not very effective in the physical world–they either completely fail to cause misclassification or only work in restricted cases where a relatively complex image is perturbed and printed on paper. In this paper we propose a new attack algorithm–Robust Physical Perturbations (RP2)– that generates perturbations by taking images under different conditions into account. Our algorithm can create spatially-constrained perturbations that mimic vandalism or art to reduce the likelihood of detection by a casual observer. We show that adversarial examples generated by RP2 achieve high success rates under various conditions for real road sign recognition by using an evaluation methodology that captures physical world conditions. We physically realized and evaluated two attacks, one that causes a Stop sign to be misclassified as a Speed Limit sign in 100% of the testing conditions, and one that causes a Right Turn sign to be misclassified as either a Stop or Added Lane sign in 100% of the testing conditions.

Uber Drivers Hacking the System to Cause Surge Pricing

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/08/uber_drivers_ha.html

Interesting story about Uber drivers who have figured out how to game the company’s algorithms to cause surge pricing:

According to the study. drivers manipulate Uber’s algorithm by logging out of the app at the same time, making it think that there is a shortage of cars.

[…]

The study said drivers have been coordinating forced surge pricing, after interviews with drivers in London and New York, and research on online forums such as Uberpeople.net. In a post on the website for drivers, seen by the researchers, one person said: “Guys, stay logged off until surge. Less supply high demand = surge.”

.

Passengers, of course, have long had tricks to avoid surge pricing.

I expect to see more of this sort of thing as algorithms become more prominent in our lives.

Hacking Slot Machines by Reverse-Engineering the Random Number Generators

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/08/hacking_slot_ma.html

Interesting story:

The venture is built on Alex’s talent for reverse engineering the algorithms — known as pseudorandom number generators, or PRNGs — that govern how slot machine games behave. Armed with this knowledge, he can predict when certain games are likeliest to spit out money­insight that he shares with a legion of field agents who do the organization’s grunt work.

These agents roam casinos from Poland to Macau to Peru in search of slots whose PRNGs have been deciphered by Alex. They use phones to record video of a vulnerable machine in action, then transmit the footage to an office in St. Petersburg. There, Alex and his assistants analyze the video to determine when the games’ odds will briefly tilt against the house. They then send timing data to a custom app on an agent’s phone; this data causes the phones to vibrate a split second before the agent should press the “Spin” button. By using these cues to beat slots in multiple casinos, a four-person team can earn more than $250,000 a week.

It’s an interesting article; I have no idea how much of it is true.

The sad part is that the slot-machine vulnerability is so easy to fix. Although the article says that “writing such algorithms requires tremendous mathematical skill,” it’s really only true that designing the algorithms requires that skill. Using any secure encryption algorithm or hash function as a PRNG is trivially easy. And there’s no reason why the system can’t be designed with a real RNG. There is some randomness in the system somewhere, and it can be added into the mix as well. The programmers can use a well-designed algorithm, like my own Fortuna, but even something less well-thought-out is likely to foil this attack.

Landmine-clearing Pi-powered C-Turtle

Post Syndicated from Janina Ander original https://www.raspberrypi.org/blog/landmine-c-turtle/

In an effort to create a robot that can teach itself to navigate different terrains, scientists at Arizona State University have built C-Turtle, a Raspberry Pi-powered autonomous cardboard robot with turtle flippers. This is excellent news for people who live in areas with landmines: C-Turtle is a great alternative to current landmine-clearing robots, since it is much cheaper, and much easier to assemble.

C-Turtle ASU

Photo by Charlie Leight/ASU Now

Why turtle flippers?

As any user of Python will tell you*, turtles are amazing. Moreover, as the evolutionary biologist of the C-Turtle team, Andrew Jansen, will tell you, considering their bulk** turtles move very well on land with the help of their flippers. Consequently, the team tried out prototypes with cardboard flippers imitating the shape of turtle flippers. Then they compared their performance to that of prototypes with rectangular or oval ‘flippers’. And 157 million years of evolution*** won out: the robots with turtle flippers were best at moving forward.

C-Turtle ASU

Field testing with Assistant Professor Heni Ben Amor, one of the C-Turtle team’s leaders (Photo by Charlie Leight/ASU Now)

If it walks like a C-Turtle…

But the scientists didn’t just slap turtle flippers on their robot and then tell it to move like a turtle! Instead, they implemented machine learning algorithms on the Pi Zero that serves as C-Turtle’s brain, and then simply let the robot do its thing. Left to its own devices, it used the reward and punishment mechanisms of its algorithms to learn the most optimal way of propelling itself forward. And lo and behold, C-Turtle taught itself to move just like a live turtle does!

Robotic C-Turtle

This is “Robotic C-Turtle” by ASU Now on Vimeo, the home for high quality videos and the people who love them.

Landmine clearance with C-Turtle

Robots currently used to clear landmines are very expensive, since they are built to withstand multiple mine explosions. Conversely, the total cost of C-Turtle comes to about $70 (~£50) – that’s cheap enough to make it disposable. It is also more easily assembled, it doesn’t need to be remotely controlled, and it can learn to navigate new terrains. All this makes it perfect for clearing minefields.

BBC Click on Twitter

Meet C-Turtle, the landmine detecting robot. VIDEO https://t.co/Kjc6WxRC8I

C-Turtles in space?****

The researchers hope that robots similar to C-Turtle can used for space exploration. They found that the C-Turtle prototypes that had performed very well in the sandpits in their lab didn’t really do as well when they were released in actual desert conditions. By analogy, robots optimized for simulated planetary conditions might not actually perform well on-site. The ASU scientists imagine that C-Turtle materials and a laser cutter for the cardboard body could be carried on board a Mars mission. Then Martian C-Turtle design could be optimized after landing, and the robot could teach itself how best to navigate real Martian terrain.

There are already Raspberry Pis in space – imagine if they actually made it to Mars! Dave would never recover

Congrats to Assistant Professors Heni Ben Amor and Daniel Aukes, and to the rest of the C-Turtle team, on their achievement! We at Pi Towers are proud that our little computer is part of this amazing project.

C-Turtle ASU

Photo by Charlie Leight/ASU Now

* Check out our Turtley amazing resource to find out why!

** At a length of 7ft, leatherback sea turtles can weigh 1,500lb!

*** That’s right: turtles survived the extinction of the dinosaurs!

**** Is anyone else thinking of Great A’Tuin right now? Anyone? Just me? Oh well.

The post Landmine-clearing Pi-powered C-Turtle appeared first on Raspberry Pi.

US Army Researching Bot Swarms

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/07/us_army_researc.html

The US Army Research Agency is funding research into autonomous bot swarms. From the announcement:

The objective of this CRA is to perform enabling basic and applied research to extend the reach, situational awareness, and operational effectiveness of large heterogeneous teams of intelligent systems and Soldiers against dynamic threats in complex and contested environments and provide technical and operational superiority through fast, intelligent, resilient and collaborative behaviors. To achieve this, ARL is requesting proposals that address three key Research Areas (RAs):

RA1: Distributed Intelligence: Establish the theoretical foundations of multi-faceted distributed networked intelligent systems combining autonomous agents, sensors, tactical super-computing, knowledge bases in the tactical cloud, and human experts to acquire and apply knowledge to affect and inform decisions of the collective team.

RA2: Heterogeneous Group Control: Develop theory and algorithms for control of large autonomous teams with varying levels of heterogeneity and modularity across sensing, computing, platforms, and degree of autonomy.

RA3: Adaptive and Resilient Behaviors: Develop theory and experimental methods for heterogeneous teams to carry out tasks under the dynamic and varying conditions in the physical world.

Slashdot thread.

And while we’re on the subject, this is an excellent report on AI and national security.

Google Removes Torrent Sites From ‘Results Carousel’

Post Syndicated from Ernesto original https://torrentfreak.com/google-removes-torrent-sites-from-results-carousel-100722/

Two weeks ago we noticed a ‘handy’ feature where Google highlighted various torrent sites in its search results.

People who typed “best torrent sites” into the search box would see a reel of popular sites such as The Pirate Bay and RARBG in the results, featured with their official logos and all.

Google employees obviously didn’t curate the list themselves. They are a Google feature called the “results carousel,” which is generated based on an algorithm. Still, considering the constant criticism the search engine faces from rightsholders, it’s a sensitive topic.

The torrent site carousel

It appears that the search engine itself wasn’t very happy with the featured search results either. This week, the torrent sites were quietly banned from the search carousel feature. According to the company, it wasn’t working as intended.

“We have investigated this particular issue and determined that this results carousel wasn’t working in the intended manner, and we have now fixed the issue,” a Google spokesperson informed TorrentFreak.

Although Google carefully avoids the words copyright and piracy in its comments, it’s quite obvious what motivated this decision. The company doesn’t want to highlight any pirate sites, to avoid yet another copyright controversy.

That the intervention was triggered by “piracy” concerns is backed up by another change. While various “streaming sites” are still prominently listed in a search carousel, the pirate sites were carefully stripped from there as well.

A few days ago it still listed sites including Putlocker, Alluc, and Movie4k.to, but only legitimate streaming portals remain on the list today. That change definitely required some human intervention.

Only ‘legitimate’ streaming postals now

This isn’t the first time that Google’s “rich” search results have featured pirate sites. The same thing happened in the past when the search engine displayed pirate site ratings of movies, next to ratings from regular review sites such as IMDb and Rotten Tomatoes.

Apparently, Google’s search engine algorithms need some anti-piracy fine-tuning, every now and then.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.

Amazon Redshift Spectrum Extends Data Warehousing Out to Exabytes—No Loading Required

Post Syndicated from Maor Kleider original https://aws.amazon.com/blogs/big-data/amazon-redshift-spectrum-extends-data-warehousing-out-to-exabytes-no-loading-required/

When we first looked into the possibility of building a cloud-based data warehouse many years ago, we were struck by the fact that our customers were storing ever-increasing amounts of data, and yet only a small fraction of that data ever made it into a data warehouse or Hadoop system for analysis. We saw that this wasn’t just a cloud-specific anomaly. It was also true in the broader industry, where the growth rate of the enterprise storage market segment greatly surpassed that of the data warehousing market segment.

We dubbed this the “dark data” problem. Our customers knew that there was untapped value in the data they collected; why else would they spend money to store it? But the systems available to them to analyze this data were simply too slow, complex, and expensive for them to use on all but a select subset of this data. They were storing it with optimistic hope that, someday, someone would find a solution.

Amazon Redshift became one of the fastest-growing AWS services because it helped solve the dark data problem. It was at least an order of magnitude less expensive and faster than most alternatives available. And Amazon Redshift was fully managed from the start—you didn’t have to worry about capacity, provisioning, patching, monitoring, backups, and a host of other DBA headaches. Many customers, including Vevo, Yelp, Redfin, and Edmunds, migrated to Amazon Redshift to improve query performance, reduce DBA overhead, and lower the cost of analytics.

And our customers’ data continues to grow at a very fast rate. Across the board, gigabytes to petabytes, the average Amazon Redshift customer doubles the data analyzed every year. That’s why we implement features that help customers handle their growing data, for example to double the query throughput or improve the compression ratios from 3x to 4x. That gives our customers some time before they have to consider throwing away data or removing it from their analytic environments. However, there is an increasing number of AWS customers who each generate a petabyte of data every day—that’s an exabyte in only three years. There wasn’t a solution for customers like that. If your data is doubling every year, it’s not long before you have to find new, disruptive approaches that transform the cost, performance, and simplicity curves for managing data.

Let’s look at the options available today. You can use Hadoop-based technologies like Apache Hive with Amazon EMR. This is actually a pretty great solution because it makes it easy and cost-effective to operate directly on data in Amazon S3 without ingestion or transformation. You can spin up clusters as you wish when you need, and size them right for that specific job you’re running. These systems are great at high scale-out processing like scans, filters, and aggregates. On the other hand, they’re not that good at complex query processing. For example, join processing requires data to be shuffled across nodes—for a large amount of data and large numbers of nodes that gets very slow. And joins are intrinsic to any meaningful analytics problem.

You can also use a columnar MPP data warehouse like Amazon Redshift. These systems make it simple to run complex analytic queries with orders of magnitude faster performance for joins and aggregations performed over large datasets. Amazon Redshift, in particular, leverages high-performance local disks, sophisticated query execution. and join-optimized data formats. Because it is just standard SQL, you can keep using your existing ETL and BI tools. But you do have to load data, and you have to provision clusters against the storage and CPU requirements you need.

Both solutions have powerful attributes, but they force you to choose which attributes you want. We see this as a “tyranny of OR.” You can have the throughput of local disks OR the scale of Amazon S3. You can have sophisticated query optimization OR high-scale data processing. You can have fast join performance with optimized formats OR a range of data processing engines that work against common data formats. But you shouldn’t have to choose. At this scale, you really can’t afford to choose. You need “all of the above.”

Redshift Spectrum

We built Redshift Spectrum to end this “tyranny of OR.” With Redshift Spectrum, Amazon Redshift customers can easily query their data in Amazon S3. Like Amazon EMR, you get the benefits of open data formats and inexpensive storage, and you can scale out to thousands of nodes to pull data, filter, project, aggregate, group, and sort. Like Amazon Athena, Redshift Spectrum is serverless and there’s nothing to provision or manage. You just pay for the resources you consume for the duration of your Redshift Spectrum query. Like Amazon Redshift itself, you get the benefits of a sophisticated query optimizer, fast access to data on local disks, and standard SQL. And like nothing else, Redshift Spectrum can execute highly sophisticated queries against an exabyte of data or more—in just minutes.

Redshift Spectrum is a built-in feature of Amazon Redshift, and your existing queries and BI tools will continue to work seamlessly. Under the covers, we manage a fleet of thousands of Redshift Spectrum nodes spread across multiple Availability Zones. These are transparently scaled and allocated to your queries based on the data that you need to process, with no provisioning or commitments. Redshift Spectrum is also highly concurrent—you can access your Amazon S3 data from any number of Amazon Redshift clusters.

The life of a Redshift Spectrum query

It all starts when Redshift Spectrum queries are submitted to the leader node of your Amazon Redshift cluster. The leader node optimizes, compiles, and pushes the query execution to the compute nodes in your Amazon Redshift cluster. Next, the compute nodes obtain the information describing the external tables from your data catalog, dynamically pruning nonrelevant partitions based on the filters and joins in your queries. The compute nodes also examine the data available locally and push down predicates to efficiently scan only the relevant objects in Amazon S3.

The Amazon Redshift compute nodes then generate multiple requests depending on the number of objects that need to be processed, and submit them concurrently to Redshift Spectrum, which pools thousands of Amazon EC2 instances per AWS Region. The Redshift Spectrum worker nodes scan, filter, and aggregate your data from Amazon S3, streaming required data for processing back to your Amazon Redshift cluster. Then, the final join and merge operations are performed locally in your cluster and the results are returned to your client.

Redshift Spectrum’s architecture offers several advantages. First, it elastically scales compute resources separately from the storage layer in Amazon S3. Second, it offers significantly higher concurrency because you can run multiple Amazon Redshift clusters and query the same data in Amazon S3. Third, Redshift Spectrum leverages the Amazon Redshift query optimizer to generate efficient query plans, even for complex queries with multi-table joins and window functions. Fourth, it operates directly on your source data in its native format (Parquet, RCFile, CSV, TSV, Sequence, Avro, RegexSerDe and more to come soon). This means that no data loading or transformation is needed. This also eliminates data duplication and associated costs. Fifth, operating on open data formats gives you the flexibility to leverage other AWS services and execution engines across your various teams to collaborate on the same data in Amazon S3. You get all of this, and because Redshift Spectrum is a feature of Amazon Redshift, you get the same level of end-to-end security, compliance, and certifications as with Amazon Redshift.

Designed for performance and cost-effectiveness

With Amazon Redshift Spectrum, you pay only for the queries you run against the data that you actually scan. We encourage you to leverage file partitioning, columnar data formats, and data compression to significantly minimize the amount of data scanned in Amazon S3. This is important for data warehousing because it dramatically improves query performance and reduces cost. Partitioning your data in Amazon S3 by date, time, or any other custom keys enables Redshift Spectrum to dynamically prune nonrelevant partitions to minimize the amount of data processed. If you store data in a columnar format, such as Parquet, Redshift Spectrum scans only the columns needed by your query, rather than processing entire rows. Similarly, if you compress your data using one of Redshift Spectrum’s supported compression algorithms, less data is scanned.

Amazon Redshift and Redshift Spectrum give you the best of both worlds. If you need to run frequent queries on the same data, you can normalize it, store it in Amazon Redshift, and get all of the benefits of a fully featured data warehouse for storing and querying structured data at a flat rate. At the same time, you can keep your additional data in multiple open file formats in Amazon S3, whether it is historical data or the most recent data, and extend your Amazon Redshift queries across your Amazon S3 data lake.

And that is how Amazon Redshift Spectrum scales data warehousing to exabytes—with no loading required. Redshift Spectrum ends the “tyranny of OR,” enabling you to store your data where you want, in the format you want, and have it available for fast processing using standard SQL when you need it, now and in the future.


Additional Reading

10 Best Practices for Amazon Redshift Spectrum
Amazon QuickSight Adds Support for Amazon Redshift Spectrum
Amazon Redshift Spectrum – Exabyte-Scale In-Place Queries of S3 Data

 


 

About the Author

Maor Kleider is a Senior Product Manager for Amazon Redshift, a fast, simple and cost-effective data warehouse. Maor is passionate about collaborating with customers and partners, learning about their unique big data use cases and making their experience even better. In his spare time, Maor enjoys traveling and exploring new restaurants with his family.

 

 

 

How to Configure Even Stronger Password Policies to Help Meet Your Security Standards by Using AWS Directory Service for Microsoft Active Directory

Post Syndicated from Ravi Turlapati original https://aws.amazon.com/blogs/security/how-to-configure-even-stronger-password-policies-to-help-meet-your-security-standards-by-using-aws-directory-service-for-microsoft-active-directory/

With AWS Directory Service for Microsoft Active Directory (Enterprise Edition), also known as AWS Microsoft AD, you can now create and enforce custom password policies for your Microsoft Windows users. AWS Microsoft AD now includes five empty password policies that you can edit and apply with standard Microsoft password policy tools such as Active Directory Administrative Center (ADAC). With this capability, you are no longer limited to the default Windows password policy. Now, you can configure even stronger password policies and define lockout policies that specify when to lock out an account after login failures.

In this blog post, I demonstrate how to edit these new password policies to help you meet your security standards by using AWS Microsoft AD. I also introduce the password attributes you can modify and demonstrate how to apply password policies to user groups in your domain.

Prerequisites

The instructions in this post assume that you already have the following components running:

  • An active AWS Microsoft AD directory.
  • An Amazon EC2 for Windows Server instance that is domain joined to your AWS Microsoft AD directory and on which you have installed ADAC.

If you still need to meet these prerequisites before proceeding:

Scenario overview

Let’s say I am the Active Directory (AD) administrator of Example Corp. At Example Corp., we have a group of technical administrators, several groups of senior managers, and general, nontechnical employees. I need to create password policies for these groups that match our security standards.

Our general employees have access only to low-sensitivity information. However, our senior managers regularly access confidential information and we want to enforce password complexity (a mix of upper and lower case letters, numbers, and special characters) to reduce the risk of data theft. For our administrators, we want to enforce password complexity policies to prevent unauthorized access to our system administration tools.

Our security standards call for the following enforced password and account lockout policies:

  • General employees – To make it easier for nontechnical general employees to remember their passwords, we do not enforce password complexity. However, we want to enforce a minimum password length of 8 characters and a lockout policy after 6 failed login attempts as a minimum bar to protect against unwanted access to our low-sensitivity information. If a general employee forgets their password and becomes locked out, we let them try again in 5 minutes, rather than require escalated password resets. We also want general employees to rotate their passwords every 60 days with no duplicated passwords in the past 10 password changes.
  • Senior managers – For senior managers, we enforce a minimum password length of 10 characters and require password complexity. An account lockout is enforced after 6 failed attempts with an account lockout duration of 15 minutes. Senior managers must rotate their passwords every 45 days, and they cannot duplicate passwords from the past 20 changes.
  • Administrators – For administrators, we enforce password complexity with a minimum password length of 15 characters. We also want to lock out accounts after 6 failed attempts, have password rotation every 30 days, and disallow duplicate passwords in the past 30 changes. When a lockout occurs, we require a special administrator to intervene and unlock the account so that we can be aware of any potential hacking.
  • Fine-Grained Password Policy administrators – To ensure that only trusted administrators unlock accounts, we have two special administrator accounts (admin and midas) that can unlock accounts. These two accounts have the same policy as the other administrators except they have an account lockout duration of 15 minutes, rather than requiring a password reset. These two accounts are also the accounts used to manage Example Corp.’s password policies.

The following table summarizes how I edit each of the four policies I intend to use.

Policy name EXAMPLE-PSO-01 EXAMPLE-PSO-02 EXAMPLE-PSO-03 EXAMPLE-PSO-05
Precedence 10 20 30 50
User group Fine-Grained Password Policy Administrators Other Administrators Senior Managers General Employees
Minimum password length 15 15 10 8
Password complexity Enable Enable Enable Disable
Maximum password age 30 days 30 days 45 days 60 days
Account complexity Enable Enable Enable Disable
Number of failed logon attempts allowed 6 6 6 6
Duration 15 minutes Not applicable 15 minutes 5 minutes
Password history 24 30 20 10
Until admin manually unlocks account Not applicable Selected Not applicable Not applicable

To implement these password policies, I use 4 of the 5 new password policies available in AWS Microsoft AD:

  1. I first explain how to configure the password policies.
  2. I then demonstrate how to apply the four password policies that match Example Corp.’s security standards for these user groups.

1. Configure password policies in AWS Microsoft AD

To help you get started with password policies, AWS has added the Fine-Grained Pwd Policy Admins AD security group to your AWS Microsoft AD directory. Any user or other security group that is part of the Fine-Grained Pwd Policy Admins group has permissions to edit and apply the five new password policies. By default, your directory Admin is part of the new group and can add other users or groups to this group.

Adding users to the Fine-Grained Pwd Policy Admins user group

Follow these steps to add more users or AD security groups to the Fine-Grained Pwd Policy Admins security group so that they can administer fine-grained password policies:

  1. Launch ADAC from your managed instance.
  2. Switch to the Tree View and navigate to CORP > Users.
  3. Find the Fine Grained Pwd Policy Admins user group. Add any users or groups in your domain to this group.

Edit password policies

To edit fine-grained password policies, open ADAC from any management instance joined to your domain. Switch to the Tree View and navigate to System > Password Settings Container. You will see the five policies containing the string -PSO- that AWS added to your directory, as shown in the following screenshot. Select a policy to edit it.

Screenshot showing the five new password policies

After editing the password policy, apply the policy by adding users or AD security groups to these policies by choosing Add. The default domain GPO applies if you do not configure any of the five password policies. For additional details about using Password Settings Container, go to Step-by-Step: Enabling and Using Fine-Grained Password Policies in AD on the Microsoft TechNet Blog.

The password attributes you can edit

AWS allows you to edit all of the password attributes except Precedence (I explain more about Precedence in the next section). These attributes include:

  • Password history
  • Minimum password length
  • Minimum password age
  • Maximum password age
  • Store password using reversible encryption
  • Password must meet complexity requirements

You also can enforce the following attributes for account lockout settings:

  • The number of failed login attempts allowed
  • Account lockout duration
  • Reset failed login attempts after a specified duration

For more details about how these attributes affect password enforcement, see AD DS: Fine-Grained Password Policies on Microsoft TechNet.

Understanding password policy precedence

AD password policies have a precedence (a numerical attribute that AD uses to determine the resultant policy) associated with them. Policies with a lower value for Precedence have higher priority than other policies. A user inherits all policies that you apply directly to the user or to any groups to which the user belongs. For example, suppose jsmith is a member of the HR group and also a member of the MANAGERS group. If I apply a policy with a Precedence of 50 to the HR group and a policy with a Precedence of 40 to MANAGERS, the policy with the Precedence value of 40 ranks higher and AD applies that policy to jsmith.

If you apply multiple policies to a user or group, the resultant policy is determined as follows by AD:

  1. If you apply a policy directly to a user, AD enforces the lowest directly applied password policy.
  2. If you did not apply a policy directly to the user, AD enforces the policy with the lowest Precedence value of all policies inherited by the user through the user’s group membership.

For more information about AD fine-grained policies, see AD DS: Fine-Grained Password Policies on Microsoft TechNet.

2. Apply password policies to user groups

In this section, I demonstrate how to apply Example Corp.’s password policies. Except in rare cases, I only apply policies by group membership, which ensures that AD does not enforce a lower priority policy on an individual user if have I added them to a group with a higher priority policy.

Because my directory is new, I use a Remote Desktop Protocol (RDP) connection to sign in to the Windows Server instance I domain joined to my AWS Microsoft AD directory. Signing in with the admin account, I launch ADAC to perform the following tasks:

  1. First, I set up my groups so that I can apply password policies to them. Later, I can create user accounts and add them to my groups and AD applies the right policy by using the policy precedence and resultant policy algorithms I discussed previously. I start by adding the two special administrative accounts (admin and midas) that I described previously to the Fine-Grained Pwd Policy Admins. Because AWS Microsoft AD adds my default admin account to Fine-Grained Pwd Policy Admins, I only need to create midas and then add midas to the Fine-Grained Pwd Policy Admins group.
  2. Next, I create the Other Administrators, Senior Managers, and General Employees groups that I described previously, as shown in the following screenshot.
    Screenshot of the groups created

For this post’s example, I use these four policies:

  1. EXAMPLE-PSO-01 (highest priority policy) – For the administrators who manage Example Corp.’s password policies. Applying this highest priority policy to the Fine-Grained Pwd Policy Admins group prevents these users from being locked out if they also are assigned to a different policy.
  2. EXAMPLE-PSO-02 (the second highest priority policy) – For Example Corp.’s other administrators.
  3. EXAMPLE-PSO-03 (the third highest priority policy) – For Example Corp.’s senior managers.
  4. EXAMPLE-PSO-05 (the lowest priority policy) – For Example Corp.’s general employees.

This leaves me one password policy (EXAMPLE-PSO-04) that I can use for in the future if needed.

I start by editing the policy, EXAMPLE-PSO-01. To edit the policy, I follow the Edit password policies section from earlier in this post. When finished, I add the Fine-Grained Pwd Policy Admins group to that policy, as shown in the following screenshot. I then repeat the process for each of the remaining policies, as described in the Scenario overview section earlier in this post.

Screenshot of adding the Fine-Grained Pwd Policy Admins group to the EXAMPLE-PSO-01 policy

Though AD enforces new password policies, the timing related to how password policies replicate in the directory, the types of attributes that are changed, and the timing of user password changes can cause variability in the immediacy of policy enforcement. In general, after the policies are replicated throughout the directory, attributes that affect account lockout and password age take effect. Attributes that affect the quality of a password, such as password length, take effect when the password is changed. If the password age for a user is in compliance, but their password strength is out of compliance, the user is not forced to change their password. For more information password policy impact, see this Microsoft TechNet article.

Summary

In this post, I have demonstrated how you can configure strong password policies to meet your security standards by using AWS Microsoft AD. To learn more about AWS Microsoft AD, see the AWS Directory Service home page.

If you have comments about this post, submit them in the “Comments” section below. If you have questions about this blog post, start a new thread on the Directory Service forum.

– Ravi

Handy: Google Highlights ‘Best Torrent Sites’ in Search Results

Post Syndicated from Ernesto original https://torrentfreak.com/handy-google-highlights-best-torrent-sites-in-search-results-170709/

With torrent sites dropping like flies recently, a lot of people are looking for alternatives.

For many, Google is the preferred choice to find them, and the search engine is actually quite helpful.

When you type in “best torrent sites” or just “torrent sites,” Google.com provides a fancy reel of several high traffic indexers.

The search engine displays the names of sites such as RARBG, The Pirate Bay and 1337x as well as their logo. When you click on this link, Google brings up all results for the associated term.

While it’s a thought provoking idea to think that Google employees are manually curating the list, the entire process is likely automated. Still, many casual torrent users might find it quite handy. Whether rightsholders will be equally excited is another question though.

The automated nature of this type of search result display also creates another problem. While many people know that most torrent sites offer pirated content, this is quite different with streaming portals.

This leads to a confusing situation where Google lists both legal and unauthorized streaming platforms when users search for “streaming sites.”

The screenshot below shows the pirate streaming site Putlocker next to Hulu and Crackle. The same lineup also rotates various other pirate sites such as Alluc and Movie4k.to.

The reels in question are most likely generated by algorithms, which don’t distinguish between authorized and unauthorized sources. Still, given the repeated criticism Hollywood has for Google for its supposed facilitation of piracy, it’s a bit unfortunate, to say the least.

This isn’t the first time that Google’s “rich” search results have featured pirate sites. The same happened in the past when the search engine displayed pirate site ratings of movies, next to ratings from regular review sites such as IMDb and Rotten Tomatoes.

We can expect the MPAA and others to take note, and bring these and other issues up at their convenience.

Note: the search reel doesn’t appear on many localized Google domains. We tested and confirmed it only on Google.com.

Source: TF, for the latest info on copyright, file-sharing, torrent sites and ANONYMOUS VPN services.

Perform Near Real-time Analytics on Streaming Data with Amazon Kinesis and Amazon Elasticsearch Service

Post Syndicated from Tristan Li original https://aws.amazon.com/blogs/big-data/perform-near-real-time-analytics-on-streaming-data-with-amazon-kinesis-and-amazon-elasticsearch-service/

Nowadays, streaming data is seen and used everywhere—from social networks, to mobile and web applications, IoT devices, instrumentation in data centers, and many other sources. As the speed and volume of this type of data increases, the need to perform data analysis in real time with machine learning algorithms and extract a deeper understanding from the data becomes ever more important. For example, you might want a continuous monitoring system to detect sentiment changes in a social media feed so that you can react to the sentiment in near real time.

In this post, we use Amazon Kinesis Streams to collect and store streaming data. We then use Amazon Kinesis Analytics to process and analyze the streaming data continuously. Specifically, we use the Kinesis Analytics built-in RANDOM_CUT_FOREST function, a machine learning algorithm, to detect anomalies in the streaming data. Finally, we use Amazon Kinesis Firehose to export the anomalies data to Amazon Elasticsearch Service (Amazon ES). We then build a simple dashboard in the open source tool Kibana to visualize the result.

Solution overview

The following diagram depicts a high-level overview of this solution.

Amazon Kinesis Streams

You can use Amazon Kinesis Streams to build your own streaming application. This application can process and analyze streaming data by continuously capturing and storing terabytes of data per hour from hundreds of thousands of sources.

Amazon Kinesis Analytics

Kinesis Analytics provides an easy and familiar standard SQL language to analyze streaming data in real time. One of its most powerful features is that there are no new languages, processing frameworks, or complex machine learning algorithms that you need to learn.

Amazon Kinesis Firehose

Kinesis Firehose is the easiest way to load streaming data into AWS. It can capture, transform, and load streaming data into Amazon S3, Amazon Redshift, and Amazon Elasticsearch Service.

Amazon Elasticsearch Service

Amazon ES is a fully managed service that makes it easy to deploy, operate, and scale Elasticsearch for log analytics, full text search, application monitoring, and more.

Solution summary

The following is a quick walkthrough of the solution that’s presented in the diagram:

  1. IoT sensors send streaming data into Kinesis Streams. In this post, you use a Python script to simulate an IoT temperature sensor device that sends the streaming data.
  2. By using the built-in RANDOM_CUT_FOREST function in Kinesis Analytics, you can detect anomalies in real time with the sensor data that is stored in Kinesis Streams. RANDOM_CUT_FOREST is also an appropriate algorithm for many other kinds of anomaly-detection use cases—for example, the media sentiment example mentioned earlier in this post.
  3. The processed anomaly data is then loaded into the Kinesis Firehose delivery stream.
  4. By using the built-in integration that Kinesis Firehose has with Amazon ES, you can easily export the processed anomaly data into the service and visualize it with Kibana.

Implementation steps

The following sections walk through the implementation steps in detail.

Creating the delivery stream

  1. Open the Amazon Kinesis Streams console.
  2. Create a new Kinesis stream. Give it a name that indicates it’s for raw incoming stream data—for example, RawStreamData. For Number of shards, type 1.
  3. The Python code provided below simulates a streaming application, such as an IoT device, and generates random data and anomalies into a Kinesis stream. The code generates two temperature ranges, where the first range is the hypothetical sensor’s normal operating temperature range (10–20), and the second is the anomaly temperature range (100–120).Make sure to change the stream name on line 16 and 20 and the Region on line 6 to match your configuration. Alternatively, you can download the Amazon Kinesis Data Generator from this repository and use it to generate the data.
    import json
    import datetime
    import random
    import testdata
    from boto import kinesis
    
    kinesis = kinesis.connect_to_region("us-east-1")
    
    def getData(iotName, lowVal, highVal):
       data = {}
       data["iotName"] = iotName
       data["iotValue"] = random.randint(lowVal, highVal) 
       return data
    
    while 1:
       rnd = random.random()
       if (rnd < 0.01):
          data = json.dumps(getData("DemoSensor", 100, 120))  
          kinesis.put_record("RawStreamData", data, "DemoSensor")
          print '***************************** anomaly ************************* ' + data
       else:
          data = json.dumps(getData("DemoSensor", 10, 20))  
          kinesis.put_record("RawStreamData", data, "DemoSensor")
          print data

  4. Open the Amazon Elasticsearch Service console and create a new domain.
    1. Give the domain a unique name. In the Configure cluster screen, use the default settings.
    2. In the Set up access policy screen, in the Set the domain access policy list, choose Allow access to the domain from specific IP(s).
    3. Enter the public IP address of your computer.
      Note: If you’re working behind a proxy or firewall, see the “Use a proxy to simplify request signing” section in this AWS Database blog post to learn how to work with a proxy. For additional information about securing access to your Amazon ES domain, see How to Control Access to Your Amazon Elasticsearch Domain in the AWS Security Blog.
  5. After the Amazon ES domain is up and running, you can set up and configure Kinesis Firehose to export results to Amazon ES:
    1. Open the Amazon Kinesis Firehose console and choose Create Delivery Stream.
    2. In the Destination dropdown list, choose Amazon Elasticsearch Service.
    3. Type a stream name, and choose the Amazon ES domain that you created in Step 4.
    4. Provide an index name and ES type. In the S3 bucket dropdown list, choose Create New S3 bucket. Choose Next.
    5. In the configuration, change the Elasticsearch Buffer size to 1 MB and the Buffer interval to 60s. Use the default settings for all other fields. This shortens the time for the data to reach the ES cluster.
    6. Under IAM Role, choose Create/Update existing IAM role.
      The best practice is to create a new role every time. Otherwise, the console keeps adding policy documents to the same role. Eventually the size of the attached policies causes IAM to reject the role, but it does it in a non-obvious way, where the console basically quits functioning.
    7. Choose Next to move to the Review page.
  6. Review the configuration, and then choose Create Delivery Stream.
  7. Run the Python file for 1–2 minutes, and then press Ctrl+C to stop the execution. This loads some data into the stream for you to visualize in the next step.

Analyzing the data

Now it’s time to analyze the IoT streaming data using Amazon Kinesis Analytics.

  1. Open the Amazon Kinesis Analytics console and create a new application. Give the application a name, and then choose Create Application.
  2. On the next screen, choose Connect to a source. Choose the raw incoming data stream that you created earlier. (Note the stream name Source_SQL_STREAM_001 because you will need it later.)
  3. Use the default settings for everything else. When the schema discovery process is complete, it displays a success message with the formatted stream sample in a table as shown in the following screenshot. Review the data, and then choose Save and continue.
  4. Next, choose Go to SQL editor. When prompted, choose Yes, start application.
  5. Copy the following SQL code and paste it into the SQL editor window.
    CREATE OR REPLACE STREAM "TEMP_STREAM" (
       "iotName"        varchar (40),
       "iotValue"   integer,
       "ANOMALY_SCORE"  DOUBLE);
    -- Creates an output stream and defines a schema
    CREATE OR REPLACE STREAM "DESTINATION_SQL_STREAM" (
       "iotName"       varchar(40),
       "iotValue"       integer,
       "ANOMALY_SCORE"  DOUBLE,
       "created" TimeStamp);
     
    -- Compute an anomaly score for each record in the source stream
    -- using Random Cut Forest
    CREATE OR REPLACE PUMP "STREAM_PUMP_1" AS INSERT INTO "TEMP_STREAM"
    SELECT STREAM "iotName", "iotValue", ANOMALY_SCORE FROM
      TABLE(RANDOM_CUT_FOREST(
        CURSOR(SELECT STREAM * FROM "SOURCE_SQL_STREAM_001")
      )
    );
    
    -- Sort records by descending anomaly score, insert into output stream
    CREATE OR REPLACE PUMP "OUTPUT_PUMP" AS INSERT INTO "DESTINATION_SQL_STREAM"
    SELECT STREAM "iotName", "iotValue", ANOMALY_SCORE, ROWTIME FROM "TEMP_STREAM"
    ORDER BY FLOOR("TEMP_STREAM".ROWTIME TO SECOND), ANOMALY_SCORE DESC;

 

  1. Choose Save and run SQL.
    As the application is running, it displays the results as stream data arrives. If you don’t see any data coming in, run the Python script again to generate some fresh data. When there is data, it appears in a grid as shown in the following screenshot.Note that you are selecting data from the source stream name Source_SQL_STREAM_001 that you created previously. Also note the ANOMALY_SCORE column. This is the value that the Random_Cut_Forest function calculates based on the temperature ranges provided by the Python script. Higher (anomaly) temperature ranges have a higher score.Looking at the SQL code, note that the first two blocks of code create two new streams to store temporary data and the final result. The third block of code analyzes the raw source data (Stream_Pump_1) using the Random_Cut_Forest function. It calculates an anomaly score (ANOMALY_SCORE) and inserts it into the TEMP_STREAM stream. The final code block loads the result stored in the TEMP_STREAM into DESTINATION_SQL_STREAM.
  2. Choose Exit (done editing) next to the Save and run SQL button to return to the application configuration page.

Load processed data into the Kinesis Firehose delivery stream

Now, you can export the result from DESTINATION_SQL_STREAM into the Amazon Kinesis Firehose stream that you created previously.

  1. On the application configuration page, choose Connect to a destination.
  2. Choose the stream name that you created earlier, and use the default settings for everything else. Then choose Save and Continue.
  3. On the application configuration page, choose Exit to Kinesis Analytics applications to return to the Amazon Kinesis Analytics console.
  4. Run the Python script again for 4–5 minutes to generate enough data to flow through Amazon Kinesis Streams, Kinesis Analytics, Kinesis Firehose, and finally into the Amazon ES domain.
  5. Open the Kinesis Firehose console, choose the stream, and then choose the Monitoring
  6. As the processed data flows into Kinesis Firehose and Amazon ES, the metrics appear on the Delivery Stream metrics page. Keep in mind that the metrics page takes a few minutes to refresh with the latest data.
  7. Open the Amazon Elasticsearch Service dashboard in the AWS Management Console. The count in the Searchable documents column increases as shown in the following screenshot. In addition, the domain shows a cluster health of Yellow. This is because, by default, it needs two instances to deploy redundant copies of the index. To fix this, you can deploy two instances instead of one.

Visualize the data using Kibana

Now it’s time to launch Kibana and visualize the data.

  1. Use the ES domain link to go to the cluster detail page, and then choose the Kibana link as shown in the following screenshot.

    If you’re working behind a proxy or firewall, see the “Use a proxy to simplify request signing” section in this blog post to learn how to work with a proxy.
  2. In the Kibana dashboard, choose the Discover tab to perform a query.
  3. You can also visualize the data using the different types of charts offered by Kibana. For example, by going to the Visualize tab, you can quickly create a split bar chart that aggregates by ANOMALY_SCORE per minute.


Conclusion

In this post, you learned how to use Amazon Kinesis to collect, process, and analyze real-time streaming data, and then export the results to Amazon ES for analysis and visualization with Kibana. If you have comments about this post, add them to the “Comments” section below. If you have questions or issues with implementing this solution, please open a new thread on the Amazon Kinesis or Amazon ES discussion forums.


Next Steps

Take your skills to the next level. Learn real-time clickstream anomaly detection with Amazon Kinesis Analytics.

 


About the Author

Tristan Li is a Solutions Architect with Amazon Web Services. He works with enterprise customers in the US, helping them adopt cloud technology to build scalable and secure solutions on AWS.

 

 

 

 

[$] Zero-copy networking

Post Syndicated from corbet original https://lwn.net/Articles/726917/rss

In many performance-oriented settings, the number of times that data is
copied puts an upper limit on how fast things can go. As a result,
zero-copy algorithms have long been of interest, even though the benefits
achieved in practice tend to be disappointing.
Networking is often performance-sensitive and is definitely dominated by
the copying of data, so an interest in zero-copy algorithms in networking
comes naturally. A set of patches under review makes that capability
available, in some settings at least.

Latency Distribution Graph in AWS X-Ray

Post Syndicated from Randall Hunt original https://aws.amazon.com/blogs/aws/latency-distribution-graph-in-aws-x-ray/

We’re continuing to iterate on the AWS X-Ray service based on customer feedback and today we’re excited to release a set of tools to help you quickly dive deep on latencies in your applications. Visual Node and Edge latency distribution graphs are shown in a handy new “Service Details” side bar in your X-Ray Service Map.

The X-Ray service graph gives you a visual representation of services and their interactions over a period of time that you select. The nodes represent services and the edges between the nodes represent calls between the services. The nodes and edges each have a set of statistics associated with them. While the visualizations provided in the service map are useful for estimating the average latency in an application they don’t help you to dive deep on specific issues. Most of the time issues occur at statistical outliers. To alleviate this X-Ray computes histograms like the one above help you solve those 99th percentile bugs.

To see a Response Distribution for a Node just click on it in the service graph. You can also click on the edges between the nodes to see the Response Distribution from the viewpoint of the calling service.

The team had a few interesting problems to solve while building out this feature and I wanted to share a bit of that with you now! Given the large number of traces an app can produce it’s not a great idea (for your browser) to plot every single trace client side. Instead most plotting libraries, when dealing with many points, use approximations and bucketing to get a network and performance friendly histogram. If you’ve used monitoring software in the past you’ve probably seen as you zoom in on the data you get higher fidelity. The interesting thing about the latencies coming in from X-Ray is that they vary by several orders of magnitude.

If the latencies were distributed between strictly 0s and 1s you could easily just create 10 buckets of 100 milliseconds. If your apps are anything like mine there’s a lot of interesting stuff happening in the outliers, so it’s beneficial to have more fidelity at 1% and 99% than it is at 50%. The problem with fixed bucket sizes is that they’re not necessarily giving you an accurate summary of data. So X-Ray, for now, uses dynamic bucket sizing based on the t-digests algorithm by Ted Dunning and Otmar Ertl. One of the distinct advantages of this algorithm over other approximation algorithms is its accuracy and precision at extremes (where most errors typically are).

An additional advantage of X-Ray over other monitoring software is the ability to measure two perspectives of latency simultaneously. Developers almost always have some view into the server side latency from their application logs but with X-Ray you can examine latency from the view of each of the clients, services, and microservices that you’re interacting with. You can even dive deeper by adding additional restrictions and queries on your selection. You can identify the specific users and clients that are having issues at that 99th percentile.

This info has already been available in API calls to GetServiceGraph as ResponseTimeHistogram but now we’re exposing it in the console as well to make it easier for customers to consume. For more information check out the documentation here.

Randall