uohzoaix / kddcup2012 Goto Github PK

In order to speed up data processing, we need to convert training data set into 
compact version.
# remote the field if it can be indicated by the other field
# store it as repeated C struct.

Randomly divide the training set into 2 set.
The first one contains 90% instance, used to train the model.
The second contains the other 10% instance, used to calculate MAE and AUC.

Randomly sample 1/1000 training data for smoking up predictor.

Implement a demo predictor output like
https://www.kddcup2012.org/c/kddcup2012-track2/forums/t/1545/calculate-auc