QC quick hits

[dfsnow]

Some quick QC tasks based on the model outputs from earlier this week:

• The RMSE of the linear model on the test set is suspiciously high, particularly in Barrington. I'm guessing there are a few non-market/erroneous sales hiding out in Barrington.
• Some PINs worth checking out from my fast QC earlier this week:
  • 11181100020000 - Seems like a gut reno in Evanston
  • 12133010680000 - Sqft too low?
  • 14322140240000 - High $/sqft
  • 05211000140000 - High $/sqft
  • 20142150160000 - High $/sqft
• Let's also check all residential PINs in ccao.pin_weird
• Do a quick histogram/scatterplot of predictions for each town, looking for enormous outliers
• Create a quick histogram of the distribution of continuous features in the training data, again looking for outliers