Now saves the trained recommender model

Author: Max

lkauto/ensemble/ensemble_builder.py

@@ -1,6 +1,10 @@
 import pandas as pd
 import numpy as np
+from lenskit import Pipeline
 from lenskit.data import Dataset
+from typing import List
+
+from lenskit.pipeline import predict_pipeline
 
 from lkauto.utils.filer import Filer
 from lkauto.utils.get_model_from_cs import get_model_from_cs
@@ -40,7 +44,8 @@ def build_ensemble(train: Dataset,
 
     es = EnsembleSelection(ensemble_size=ensemble_size, lenskit_metric=lenskit_metric, maximize_metric=maximize_metric)
     es.ensemble_fit(ensemble_x, ensemble_y)
-    es.base_models = [get_model_from_cs(cs, feedback='explicit') for cs, weight in zip(bm_cs_list, es.weights_) if weight > 0]
+    es.base_models_tmp = [get_model_from_cs(cs, feedback='explicit') for cs, weight in zip(bm_cs_list, es.weights_) if weight > 0]
+    es.base_models = models_to_pipelines(base_models=es.base_models_tmp)
     es.old_to_new_idx = {old_i: new_i for new_i, old_i in enumerate([idx for idx, weight in enumerate(es.weights_) if weight > 0])}
 
     incumbent = {"model": str(es),
@@ -51,3 +56,12 @@ def build_ensemble(train: Dataset,
                  "weights": list(es.weights_)}
 
     return es, incumbent
+
+def models_to_pipelines(base_models) -> List[Pipeline]:
+    pipelines = []
+
+    for model in base_models:
+        pipeline = predict_pipeline(model)
+        pipelines.append(pipeline)
+
+    return pipelines

lkauto/lkauto.py

@@ -219,12 +219,12 @@ def get_best_prediction_model(train: Dataset,
         logger.info('GES Ensemble Model')
     else:
         # build model from best model configuration found by SMAC
-        model = get_model_from_cs(incumbent, feedback='explicit')
+        # model = get_model_from_cs(incumbent, feedback='explicit')
         incumbent = incumbent.get_dictionary()
         logger.info('--Best Model--')
         logger.info(incumbent)
 
-    if save:
+    if save and ensemble_size == 1:
         filer.save_model(model)
         filer.save_incumbent(incumbent)
         logger.info('Saved model and incumbent to ' + filer.output_directory_path)