Week 5 at Metis

Week 5 at Metis

Feb 12, 2016    

Week 5 at Metis and we’re getting into the meat and potatoes of supervised machine learning. Here’s a recap

Supervised Machine Learning

In 1 week, we’ve learned all of the following:

  • Naive Bayes Classification
  • Maximum Likelihood Estimation and Generalized Linear Models for:
    • Bernoulli distribution
    • Binomial distribution
    • Poisson distribution
    • Logistic regression and
    • Linear regression
  • Neural Networks and Deep Learning
  • Stochastic gradient descent

Isn’t that crazy? All of them in about 7-8 40 minute lectures! We learned their basic ideas, how to derive some of the equations, and how to apply them with Scikit-Learn. But to really understand them, I’m spending as much time as I can learning about their details.

Project 3 - Classification with Supervised Learning

This is also week 2 of 3 in our 3rd project. We had a chance to solidify our question and present what we have so far to the class.

I’m working on a Kaggle classification project sponsored by BNP Paribas Cardif to come up with the best model to expedite personal payment insurance claims.

The tough part of this challenge is that all the features are anonymized. So you can’t perform intuitive feature engineering by looking at the relevant features. I’ll have to actually study the features themselves and find the best ones to build a model with.

This would be very fun.

Sudheer Marsetti from Aetna spoke to Metis

Sudheer Marsetti from Aetna came to speak to us, and I wrote a short piece of his talk here.


We also did a bit more with SQL. Using the PostgreSQL database we built on AWS, we link our local iPython Notebooks using SQLAlchemy & Psycopg.

Resume and networking workshops

Jennifer led 2 workshops for us this week. First one on networking, and another one on our resumes and cover letters. It’s cool to learn how to customize my resume (again) but this time for a more technical role.

I’ll also be making a web friendly resume to put on this website. Coming soon

Pair Programming

Pair programming problems are getting more challenging now. We’re learning about dynamic programming and some SQL. Take a look at my work so far: Github