Nina Zumel delves into principal component regression using R (via R Bloggers):
Data tends to come from databases that must support many different tasks, so it is exactly the case that there may be columns or variables that are correlated to unknown and unwanted additional processes. The reason PCA can’t filter out these noise variables is that without use of y, standard PCA has no way of knowing what portion of the variation in each variable is important to the problem at hand and should be preserved. This can be fixed through domain knowledge (knowing which variables to use), variable pruning and y-aware scaling. Our next article will discuss these procedures; in this article we will orient ourselves with a demonstration of both what a good analysis and what a bad analysis looks like.
All the variables are also deliberately mis-scaled to model some of the difficulties of working with under-curated real world data.
This does read like an academic paper, so it’s pretty heavy reading. It’s also very good reading from a great writer, so take some time and give it a read if you do data analysis.
David Smith points out a free e-book on creating effective graphs with Microsoft R Open:
The examples were done using Microsoft R Open, but since it’s 100% compatible with R the code works with any relatively recent R version.
Naomi and Joyce presented several examples from their e-book in a recent webinar (presented by Microsoft), and fielded lots of interesting questions from the audience. If you’d like to see the recorded webinar and also receive a copy of the slides and the e-book, follow the link below to register to receive the materials via email.
The book is free, the code is available on GitHub. What more could you ask for?
In addition to the original features in the raw data, we add number of bikes rented in each of the previous 12 hours as features to provide better predictive power. We create a
computeLagFeatures()helper function to compute the 12 lag features and use it as the transformation function inrxDataStep().Note that
rxDataStep()processes data chunk by chunk and lag feature computation requires data from previous rows. IncomputLagFeatures(), we use the internal function.rxSet()to save the last n rows of a chunk to a variable lagData. When processing the next chunk, we use another internal function.rxGet()to retrieve the values stored in lagData and compute the lag features.
This is a great article for anybody wanting to dig into analytics, because they show their work.
As this is a major release, you’ll need to re-install any packages you were using (and perhaps wait a little while until package authors make any compatibility fixes needed for version 3.3.0). If you’re on the Windows platform, Tal Galili’s installr package automates the process for you. If you are using the checkpoint package (on any platform) you can simply increment the checkpoint date to anytime after May 2, 2016.
(For Microsoft R Open users, the next version to be released will be MRO 3.2.5, and MRO 3.3.0 will follow soon thereafter.)
For more information about R 3.3.0, including the detailed list of changes and bug fixes, follow the link to the announcement from the R Core Group below.
David Smith also notes that R Tools for Visual Studio 0.3 has been released:
R Tools for Visual Studio, the open-source extenstion to Visual Studio that provides an IDE for the R language, has been upgraded to include several new features.
The latest update, RTVS 0.3, now includes:
An R package manager, allowing you to review, install, and uninstall packages using a convenient user interface.
The Variable Explorer now allows you to open data-frames for viewing in an Excel workbook.
New toolbar buttons to run selected code, source the current script, import data from a URL or file, and start/stop a Shiny app.
This is a great time to get interested in R. If you’re familiar with Visual Studio, Microsoft is making great strides toward integrating things nicely.
David Smith points out a recent webinar on combining Microsoft R Server with HDInsight:
As Mario Inchiosa and Roni Burd demonstrate in this recorded webinar, Microsoft R Server can now run within HDInsight Hadoop nodes running on Microsoft Azure. Better yet, the big-data-capable algorithms of ScaleR (pdf) take advantage of the in-memory architecture of Spark, dramatically reducing the time needed to train models on large data. And if your data grows or you just need more power, you can dynamically add nodes to the HDInsight cluster using the Azure portal.
I don’t normally link to webinars (because they tend to violate my “should be viewable in a coffee break” rule of thumb) but I have a soft spot in my heart for these technologies. If you want to dig into more “mainstream” (off the Microsoft platform) Spark + R fun, check out SparkR.
Debraj GuhaThakurta, Senior Data Scientist, and Shauheen Zahirazami, Senior Machine Learning Engineer at Microsoft, demonstrate some of these capabilities in their analysis of 170M taxi trips in New York City in 2013 (about 40 Gb). Their goal was to show the use of Microsoft R Server on an HDInsight Hadoop cluster, and to that end, they created machine learning models using distributed R functions to predict (1) whether a tip was given for a taxi ride (binary classification problem), and (2) the amount of tip given (regression problem). The analyses involved building and testing different kinds of predictive models. Debraj and Shauheen uploaded the NYC Taxi data to HDFS on Azure blob storage, provisioned an HDInsight Hadoop Cluster with 2 head nodes (D12), 4 worker nodes (D12), and 1 R-server node (D4), and installed R Studio Server on the HDInsight cluster to conveniently communicate with the cluster and drive the computations from R.
To predict the tip amount, Debraj and Shauheen used linear regression on the training set (75% of the full dataset, about 127M rows). Boosted Decision Trees were used to predict whether or not a tip was paid. On the held-out test data, both models did fairly well. The linear regression model was able to predict the actual tip amount with a correlation of 0.78 (see figure below). Also, the boosted decision tree performed well on the test data with an AUC of 0.98.
If you’re looking for a data set for exploration, this is certainly a good one.
Steph Locke shows how to collapse a list of data frames into a single data table:
With my HIBPwned package, I consume the HaveIBeenPwned API and return back a list object with an element for each email address. Each element holds a data.frame of breach data or a stub response with a single column data.frame containing NA. Elements are named with the email addresses they relate to. I had a list of data.frames and I wanted a consolidated data.frame (well, I always want a data.table).
Enter data.table …
data.table has a very cool, and very fast function named
rbindlist(). This takes a list of data.frames and consolidates them into one data.table, which can, of course, be handled as a data.frame if you didn’t want to use data.table for anything else.
Something that continuously amazes me with R is just how terse the language can be without collapsing into Perl.
Andrie de Vries notes that Azure Machine Learning now supports Jupyter Notebooks with R:
I wrote about Jupyter Notebooks in September 2015 (Using R with Jupyter Notebooks), where I noted some of the great benefits of using notebooks:
Jupyter is an easy to use and convenient way of mixing code and text in the same document.
Unlike other reporting systems like RMarkdown and LaTex, Jupyter notebooks are interactive – you can run the code snippets directly in the document
This makes it easy to share and publish code samples as well as reports.
Jupyter Notebooks is a fine application, but up until now, you could only integrate it with Azure Machine Learning if you were writing Python code. This move is a big step forward for Azure ML.
Steph Locke notes that the R Consortium has agreed to support satRdays:
I’m very pleased to say that the R Consortium agreed to the support the satRday project!
The idea kicked off in November and I was over the moon with the response from the community, then we garnered support before submitting to the Consortium and I must have looped the moon a few times as we had more than 500 responses. Now the R Consortium are supporting us and we can turn all that enthusiasm into action.
This is great. I’m looking forward to this taking off and being a nice complement to SQL Saturdays in cities.
Steph Locke has created an R package to query Troy Hunt’s Have I Been Pwned? site:
The answer in life to the inevitable question of “How can I do that in R?” should be “There’s a package for that”. So when I wanted to query HaveIBeenPwned.com (HIBP) to check whether a bunch of emails had been involved in data breaches and there wasn’t an R package for HIBP, it meant that the responsibility for making one landed on my shoulders. Now, you can see if your accounts are at risk with the R package for HaveIBeenPwned.com, HIBPwned.
This is a nice confluence of two fun topics, so of course I like it.
Kevin Feasel
2016-05-17
R