Error when training with optimize_final_model = True

[chjq201410695]

My environment is ubantu 16.08.

here is my code:

column_descriptions = { 'booking': 'output', ('UN', 'TS', 'ICS', 'MS'): 'categorical' } ml_predictor = Predictor(type_of_estimator='regressor', column_descriptions=column_descriptions)

ml_predictor.train(train, optimize_final_model=True, ml_for_analytics=False, take_log_of_y=True, model_names=['LGBMRegressor', 'XGBRegressor', 'DeepLearningRegressor'], cv=9, feature_learning=False, prediction_intervals=True,compare_all_models=True)

ml_predictor.score(test, test.booking)

---

The error is:

---

ypeError Traceback (most recent call last) in () 9 take_log_of_y=True, model_names=['LGBMRegressor', 'XGBRegressor', 'DeepLearningRegressor'], 10 cv=9, feature_learning=False, ---> 11 prediction_intervals=True,compare_all_models=True) 12 13 ml_predictor.score(test, test.booking)

~/.conda/envs/ML_DNN/lib/python3.6/site-packages/auto_ml/predictor.py in train(failed resolving arguments) 668 669 # This is our main logic for how we train the final model --> 670 self.trained_final_model = self.train_ml_estimator(self.model_names, self._scorer, X_df, y) 671 672 if self.ensemble_config is not None and len(self.ensemble_config) > 0:

~/.conda/envs/ML_DNN/lib/python3.6/site-packages/auto_ml/predictor.py in train_ml_estimator(self, estimator_names, scoring, X_df, y, feature_learning, prediction_interval) 1267 self.grid_search_params = grid_search_params 1268 -> 1269 gscv_results = self.fit_grid_search(X_df, y, grid_search_params, feature_learning=feature_learning) 1270 1271 all_gs_results.append(gscv_results)

~/.conda/envs/ML_DNN/lib/python3.6/site-packages/auto_ml/predictor.py in fit_grid_search(self, X_df, y, gs_params, feature_learning, refit) 1154 generations_number=generations_number, 1155 # Do not fit the best estimator on all the data- we will do that later, possibly after increasing epochs or n_estimators -> 1156 refit=refit 1157 1158 )

TypeError: Can't instantiate abstract class EvolutionaryAlgorithmSearchCV with abstract methods _run_search

I found that the error appears when I set the optimize_final_model = True.

---

I have tried using pathos.multiprocessing.Pool instead of multiprocessing.Pool in "evolutionary_search/cv.py" mentioned by @kngo107 in issue #376 but it does not work yet.

any other advise？

[mohataher]

Can confirm, I'm receiving the same error with the following code:

ml_predictor = Predictor(type_of_estimator='regressor',  column_descriptions=column_descriptions)
ml_predictor.train(X_train, model_names=['DeepLearningRegressor'], optimize_final_model=True)
score = ml_predictor.score(X_test, X_test.y)

Error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-38-181ddefc36c8> in <module>
     14     ml_predictor = Predictor(type_of_estimator='regressor',  column_descriptions=column_descriptions)
     15     # ml_predictor.train(X_train, model_names=['DeepLearningRegressor'])
---> 16     ml_predictor.train(X_train, model_names=['DeepLearningRegressor'], optimize_final_model=True)
     17     score = ml_predictor.score(X_test, X_test.y)
     18     if (best_score<score):

c:\users\mtaher\appdata\local\programs\python\python36\lib\site-packages\auto_ml\predictor.py in train(***failed resolving arguments***)
    668 
    669         # This is our main logic for how we train the final model
--> 670         self.trained_final_model = self.train_ml_estimator(self.model_names, self._scorer, X_df, y)
    671 
    672         if self.ensemble_config is not None and len(self.ensemble_config) > 0:

c:\users\mtaher\appdata\local\programs\python\python36\lib\site-packages\auto_ml\predictor.py in train_ml_estimator(self, estimator_names, scoring, X_df, y, feature_learning, prediction_interval)
   1267                 self.grid_search_params = grid_search_params
   1268 
-> 1269                 gscv_results = self.fit_grid_search(X_df, y, grid_search_params, feature_learning=feature_learning)
   1270 
   1271                 all_gs_results.append(gscv_results)

c:\users\mtaher\appdata\local\programs\python\python36\lib\site-packages\auto_ml\predictor.py in fit_grid_search(self, X_df, y, gs_params, feature_learning, refit)
   1154                 generations_number=generations_number,
   1155                 # Do not fit the best estimator on all the data- we will do that later, possibly after increasing epochs or n_estimators
-> 1156                 refit=refit
   1157 
   1158             )

TypeError: Can't instantiate abstract class EvolutionaryAlgorithmSearchCV with abstract methods _run_search

[asvany]

workaround:

1. downgrade the scikit learn to 19.2 'pip install -U scikit-learn==0.19.2'
2. remove the dummy parameter from the constructor: auto_ml/predictor.py:1149 " pre_dispatch='1.5*n_jobs', "

((( comment from "class EvolutionaryAlgorithmSearchCV(BaseSearchCV):"

pre_dispatch : int, or string, optional
    Dummy parameter for compatibility with sklearn's GridSearch

)))