As a professional musician in my former life, the saxophone paid my rent. Thus, I thought it fitting to pay it tribute for this project, and see if I could train a convolutional neural network to detect its presence in audio clips. I created a volume of 24,000 audio samples from several hundred songs in my library, logging a record of each one in a MongoDB database at the time of creation. To label these samples, I wrote a web app using Flask and JavaScript, which I then hosted at audiosamplelabeler.com so anyone anywhere could pitch in and assist in the labeleing effort. As of 12/8/17, 2,000 out of 24,000 samples have been labeled, and the effort will continue indefinitely.

Our second project for Metis began with only two constraints:

1. Gather a substantial amount of data via web scraping
2. Build a linear regression model to attempt to predict a certain target

For our first project here at Metis, we were asked to respond to an informal RFP from a fictitious non-profit of the name Women Tech Women Yes, which solicited analytics advice to guide the placement of their promotional street teams in advance of their annual fundraising gala. As street teams would be placed at the entrances of subway stations to collect email addresses in exchange for free tickets to the gala, they suggested using the freely-available MTA data to determine which subway stations might be the most beneficial to target. The stated interest was in collecting as many signatures as possible from “…those who will attend the gala and contribute to our cause.” No specific date for the gala was given, though they mentioned it would be in early summer.