AutoML-GP5-LightAutoML-OOFs-and-Test-preds by Kaggle | Sports

About this Dataset

AutoML-GP5-LightAutoML-OOFs-and-Test-preds

[AutoML Grand Prix] 1st Place Solution Team LightAutoML testers - OOF and Test predictions

🏆 Solution description

Github repo with the training code

TLDR: 24 base regression and classification models, GBDT + NN, with their blend.

We trained all models (CatBoost and LGBM for regression, DenseLight and FT-Transformer for both regression and classification) with original and clipped target (clip all price values higher than 500k for training fold) using original and augmented with kagglex (added only in train fold) datasets (thanks to @lashfire).

Our final ensemble with 10-Fold CV scores is:

Tables

Oof Lightautomltestersmodels

@kaggle.alexryzhkov_automl_gp5_lightautoml_oofs_and_test_preds.oof_lightautomltestersmodels

38.18 MB
188533 rows
27 columns


CREATE TABLE oof_lightautomltestersmodels (
  "id" BIGINT,
  "price" DOUBLE,
  "cb_woclip_data" DOUBLE,
  "cb_woclip_wodata" DOUBLE,
  "cb_clip_data" DOUBLE,
  "cb_clip_wodata" DOUBLE,
  "lgb_woclip_data" DOUBLE,
  "lgb_woclip_wodata" DOUBLE,
  "lgb_clip_data" DOUBLE,
  "lgb_clip_wodata" DOUBLE,
  "dl_cls_woclip_data" DOUBLE,
  "dl_cls_woclip_wodata" DOUBLE,
  "dl_reg_woclip_data" DOUBLE,
  "dl_reg_woclip_wodata" DOUBLE,
  "dl_cls_clip_data" DOUBLE,
  "dl_cls_clip_wodata" DOUBLE,
  "dl_reg_clip_data" DOUBLE,
  "dl_reg_clip_wodata" DOUBLE,
  "ftt_cls_woclip_data" DOUBLE,
  "ftt_cls_woclip_wodata" DOUBLE,
  "ftt_reg_woclip_data" DOUBLE,
  "ftt_reg_woclip_wodata" DOUBLE,
  "ftt_cls_clip_data" DOUBLE,
  "ftt_cls_clip_wodata" DOUBLE,
  "ftt_reg_clip_data" DOUBLE,
  "ftt_reg_clip_wodata" DOUBLE,
  "blend" DOUBLE
);

Pred Lightautomltestersmodels

@kaggle.alexryzhkov_automl_gp5_lightautoml_oofs_and_test_preds.pred_lightautomltestersmodels

28.25 MB
125690 rows
27 columns


CREATE TABLE pred_lightautomltestersmodels (
  "id" BIGINT,
  "id_1" BIGINT,
  "cb_woclip_data" DOUBLE,
  "cb_woclip_wodata" DOUBLE,
  "cb_clip_data" DOUBLE,
  "cb_clip_wodata" DOUBLE,
  "lgb_woclip_data" DOUBLE,
  "lgb_woclip_wodata" DOUBLE,
  "lgb_clip_data" DOUBLE,
  "lgb_clip_wodata" DOUBLE,
  "dl_cls_woclip_data" DOUBLE,
  "dl_cls_woclip_wodata" DOUBLE,
  "dl_reg_woclip_data" DOUBLE,
  "dl_reg_woclip_wodata" DOUBLE,
  "dl_cls_clip_data" DOUBLE,
  "dl_cls_clip_wodata" DOUBLE,
  "dl_reg_clip_data" DOUBLE,
  "dl_reg_clip_wodata" DOUBLE,
  "ftt_cls_woclip_data" DOUBLE,
  "ftt_cls_woclip_wodata" DOUBLE,
  "ftt_reg_woclip_data" DOUBLE,
  "ftt_reg_woclip_wodata" DOUBLE,
  "ftt_cls_clip_data" DOUBLE,
  "ftt_cls_clip_wodata" DOUBLE,
  "ftt_reg_clip_data" DOUBLE,
  "ftt_reg_clip_wodata" DOUBLE,
  "blend" DOUBLE
);