Tensorflow

Almost two years ago, I completed the Machine Learning course from Stanford University on Coursera. It was a brutal course using Octave mathematical software to code, line by line, machine learning algorithms such as linear regression, neural networks, k means etc. I struggled through it and often barely passed the quizzes.

After I finished it (and after updating my Linkedin), I thought about what project I might use it for. I looked at the range of analysis that I was undertaking as an analyst with Cornerstone Capital in New York and none of it looked viable. It wasn’t the topics or the lack of data but it was the difficulty in developing a machine learning program.

Octave software meant that for each machine learning program, I had build everything from scratch. From the data input, data cleaning, initial analysis, and training and test data selection into the algorithm selection, running and prediction, it felt was extremely hard to start. Imagine a cabinetmaking job where each time you made a cabinet, you had to build all of your tools from scratch.

Mostly, it felt like no question was worth answering with machine learning if it was going to be that hard to do.

I also knew that this state would not continue. Someone would build an intuitive machine learning tool. Machine learning was too useful to let only very specific programmers be the guardians. In order for the tool to be useful, it had to be used.

About three weeks ago, I saw an ad for the Tensorflow Introduction by Coursera. $69 for a 4 week online course. It said that Tensorflow, a product from Google Deep Mind, was an intuitive machine learning interface with most of the elements already coded in. Amazing!

I did the course and am now doing my first small regression analysis using a neural network. Code available at GitHub . A very useful development!