search

thekingofkings / chicago-crime

Crime correlation anaysis
MIT License
11 stars 3 forks source link

Train and evaluate a Dynamic Coefficient Model #10

Closed thekingofkings closed 8 years ago

thekingofkings commented 8 years ago

Follow-up the issue #6 here.

Since we have already found the proper supporting case. Next step would be train and evaluate a dynamic coefficient model.

Goal

  1. Implement the dynamic coefficient model
  2. evaluate the performance
thekingofkings commented 8 years ago

Implementation is done. The evaluation is tricky next step

thekingofkings commented 8 years ago

Two way to evaluate the Dynamic Linear Regression Model

The intuitive approach is to train on one year, and use this model to predict the next year. This is, however, different from what we have in previous evaluation setting (leave-one-out).

The alternative is still leave-one-out. I hold one CA and train a model on the others. Given the new CA, I learn its coefficients w first. (Assumptions are made here)