Coming back to markets and trading (after a while), the feeling has been that the markets, and the economy as a whole, are doing good. How good? Since I haven’t been following things closely, I had to do some forensics.

Since there are already great articles on this topic(please see resources in the end) in this post a more practical way(source) is shown how I was screwing up  tests. So we are going to develop a small JAVA application and show step by step what can go wrong even without noticing it.

Aspiring data scientists and other visitors to this site often repeat the same questions. This post is the definitive collection of my answers to such questions (which may evolve over time).

This story, like all great stories, begins with a light fixture. In particular, this light fixture:

In the past few years, deep learning has seen incredible success in image recognition applications. In this post I examine how to train a convolutional neural network to recognize playing card images from a game called SET®, explore the structure of the model to get some insight into what it is “seeing”, and present a webcam application that uses the deployed model in a near-realtime setting.

In the last few years there has been an explosion in the number of courses and programs to train the next generation of data scientists. This is in response to both the perceived demand for data science (along with the rest of the buzzword soup: Big Data, the Internet of Things, Machine Learning, AI, etc.) and the perceived shortage of trained data scientists. While finding good data scientists is still a difficult problem, the increasing number of bootcamps, master’s degree programs, and online courses is starting to address this difficulty. In our experience, however, we find that many organizations are not necessarily short on data science talent, but rather that management does not properly use the talent they already have. Managers, directors, and VPs must improve their ability to work with data scientists if they want their organizations to use data more effectively; in short, they need to be more data fluent.

This post was originally published as a tutorial for DataCamp here on September 12 2017 using NetworkX 1.11. On September 20 2017, NetworkX announced the release of a new version 2.0, after two years in the making. While 2.0 introduces lots of great features (some have already been used to improve this project in postman_problems), it also introduced backwards incompatible API changes that broke the original tutorial :(. I’ve commented out lines deprecated by 2.0 and tagged with # deprecated after NX 1.11, so the changes made here are explicit. Most of the changes are around the passing and setting of attributes and return values deprecating lists for generators.

Last year, when I was on a mission to “fix GANs” I had a tendency to focus only on what the loss function is, and completely disregard the issue of how do we actually find a minimum. Here is the paper that has finally challenged that attitude:

If you have a long drive and no bluetooth or aux cord to listen to podcasts, NPR is easily the best alternative. Truck drivers agree with this statement no matter their overall views. For me, this was the case when driving home to Milwaukee from Ann Arbor where I went to a college friend’s wedding.

I’ve been telling everyone that I’d do something “data fun” when I hit 20K Twitter followers, so I posted an analysis of my podcast listeners! I used python and pandas in a Jupyter notebook for the first part, then I did a dashboard in Tableau for the last part.