MDST Ratings Analysis Challenge

Here, we'll be sharing tutorial and demo code from the MDST Ratings Analysis Challenge.

Competition Page

Scripts

ridge.py - Ridge regression baseline code. Uses regularized sum of bigram features to predict quality. Leaderboard RMSE: 1.8431

transformdata.py - Cleans up categorical variables to be used with other learners. Written by Alexander Zaitzeff

Flux

template_ridge.pbs - Use this PBS script to run the ridge regression code remotely on Flux. You'll need your account linked to our Flux allocation (contact Jonathan at stroud@umich.edu). Make sure to change this line at the top:

#PBS -M youruniqname@umich.edu

to match your actual uniqname. To run:

qsub ridge.pbs

Data

For a complete description of the dataset, see the kaggle in-class webpage:

Data Description