nonick / airbnb-booking-predict

Ансамбль народных классификаторов и плясок с бубном

airbnb-booking-predict

Score

• Single XGB on whole set of features score ndcg_5 = 0.8766 which correspondes to top 25% in private LB
  Alas, the competition is closed and this score isn't in LB.

Steps to run

• Unpack data/train_users_2_norm.csv.zip and data/test_users_norm.csv.zip
• Run train_and_predict_singleXGB.py (~15 min at 4 threads) to generate predict/xgb.csv submission data
  OR
• Run train_and_predict_stacking.py (~3 hrs at 4 threads) to generate predict/ensemble.csv submission data

Other runnables

• action_count.py (4-6 hrs at 15-20 sec per 10k entries) requires data/sessions.csv from kaggle and generates data/user_actions_count.csv
• process_input.py (~6 min) to generates data/train_users_2_norm.csv and data/test_users_norm.csv
• xgb_gridsearch.py finds best params (from described) for each classifier in ensemble