Merge branch 'main' of https://github.com/ottogin/raif_gg_hack into main

ottogin · Sep 25, 2021 · dcbfca0 · dcbfca0
2 parents dfa9172 + 8f1122b
commit dcbfca0
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 5 deletions.
diff --git a/predict.py b/predict.py
@@ -4,7 +4,7 @@
 from raif_hack.features import prepare_categorical
 from traceback import format_exc
 
-from raif_hack.model import BenchmarkModel, TwoStepBenchmarkModel, WeightedTwoStepModel
+from raif_hack.model import *
 from raif_hack.floor_processing import get_floor_nb_and_height_features
 from raif_hack.streets_reforms_processing import combine_street_region, fill_reforms_500_as_1000
 
@@ -82,7 +82,7 @@ def parse_args():
         test_df = prepare_categorical(test_df)
 
         logger.info("Load model")
-        model = WeightedTwoStepModel.load(args["mp"])
+        model = WeightedBlendModel.load(args["mp"])
         logger.info("Predict")
         test_df["per_square_meter_price"] = model.predict(
             test_df[NUM_FEATURES + CATEGORICAL_OHE_FEATURES + CATEGORICAL_STE_FEATURES]

diff --git a/raif_hack/model.py b/raif_hack/model.py
@@ -552,6 +552,8 @@ def fit(
         y_val_offer: typing.Optional[pd.Series] = None,
         X_val_manual: typing.Optional[pd.DataFrame] = None,
         y_val_manual: typing.Optional[pd.Series] = None,
+        use_best_model = False,
+        w=0.9
     ):
         y_offer, y_manual, y_val_offer, y_val_manual = np.log(y_offer), np.log(y_manual), np.log(y_val_offer), np.log(y_val_manual)
         logger.info("Fit catboost")
@@ -564,8 +566,8 @@ def fit(
         )
         X_off_prep = self.pipeline3[:-1].transform(X_manual)
         self.pipeline3[-1:].fit(X_off_prep, y_manual)
-        y_offer = 0.9*y_offer + 0.1*self.pipeline3.predict(X_offer)
-        
+        y_offer = w * y_offer + (1 - w) * self.pipeline3.predict(X_offer)
+
         X = pd.concat([X_offer, X_manual])
         y = pd.concat([y_offer, y_manual])
         WEIGHT = 0.05
@@ -587,7 +589,7 @@ def fit(
             # + CATEGORICAL_STE_FEATURES,
             # model__categorical_feature=CATEGORICAL_OHE_FEATURES
             # + CATEGORICAL_STE_FEATURES,
-            model__use_best_model=True,
+            model__use_best_model=use_best_model,
             model__eval_set=Pool(X_val_prep, y_val),
             model__sample_weight=weight,
         )