This week I’m tweeting from the @shefunilife account, an account that is switched once a week between staff and students at the University of Sheffield.

For being in RSiteCatalyst retirement, I’m ending up working on more functionality lately ¯(ツ)/¯. Here are the changes for RSiteCatalyst 1.4.8, which should be available on CRAN shortly:

April 1, 2016

The hidden layers in a neural network can be seen as different representations of the input. Do deeper layers learn “better” representations? In a network trained to solve a classification problem, this would mean that deeper layers provide better features than earlier layers. The natural hypothesis is that this is indeed the case. In this post, I test this hypothesis on an network with three hidden layers trained to classify the MNIST dataset. It is shown that deeper layers do in fact produce better representations of the input.

Batch normalization, as described in the March 2015 paper (the BN2015 paper) by Sergey Ioffe and Christian Szegedy, is a simple and effective way to improve the performance of a neural network. In the BN2015 paper, Ioffe and Szegedy show that batch normalization enables the use of higher learning rates, acts as a regularizer and can speed up training by 14 times. In this post, I show how to implement batch normalization in Tensorflow.

Renee interviews computational biologist, author, data scientist, and Michigan State PhD candidate Sebastian Raschka about how he became a data scientist, his current research, and about his book Python Machine Learning. In the audio interview, Sebastian also joins us to discuss k-fold cross-validation for our model evaluation Data Science Learning Club activity.

This past January, we (Hadley and Wes) met and discussed some of the systems challenges facing the Python and R open source communities. In particular, we wanted to explore opportunities to collaborate on tools for improving interoperability between Python, R, and external compute and storage systems.

With MLB Opening Day just over a week away, it’s time for another season of fantasy baseball. If your leagues are anything like mine, that means chains of 200+ emails hashing out every detail of league rules and setup. When are we drafting? Should we change our keeper rules? Wait, what are the keeper rules? What about changing scoring or roster settings? This goes on and on, and we usually end up changing nothing from the previous year. It’s ridiculous, but given that leagues can have a wide variety of configurations, it makes sense that the managers will almost never immediately align on an acceptable setup.

Recent advances in hardware and software have enabled the capture and the generation of different measurements of data in a wide range of fields. These measurements are generated continuously and in a very high fluctuating data rates. Examples include sensor networks, web logs, financial market, and computer network traffic. The storage, querying and processing of such data sets are highly computationally challenging tasks. Thus, there is an increasing interest on streaming architecture with some interesting outcome like Storm, Spark Streaming and Flink.