support ray tune imputer

pyzoo/zoo/automl/config/recipe.py

@@ -565,6 +565,7 @@ def search_space(self, all_available_features):
             # -------- feature related parameters
             "model": "XGBRegressor",
 
+            "imputation": tune.choice(["LastFillImpute", "FillZeroImpute"]),
             "n_estimators": self.n_estimators,
             "max_depth": self.max_depth,
             "min_child_weight": self.min_child_weight,

pyzoo/zoo/automl/impute/__init__.py

@@ -0,0 +1,15 @@
+#
+# Copyright 2018 Analytics Zoo Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#

pyzoo/zoo/automl/impute/abstract.py

@@ -0,0 +1,53 @@
+#
+# Copyright 2018 Analytics Zoo Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+from abc import ABC, abstractmethod
+
+
+class BaseImputation(ABC):
+    """
+    Abstract Base class for imputation.
+    """
+
+    @abstractmethod
+    def impute(self, input_df):
+        """
+        fill na value
+        :param input_df: input dataframe
+        :return:
+        """
+        pass
+
+    def save(self, file_path):
+        """
+        save the feature tools internal variables.
+        Some of the variables are derived after fit_transform, so only saving config is not enough.
+        :param: file_path : the file to be saved
+        :param: config: the trial config
+        :return:
+        """
+        pass
+
+    def restore(self, **config):
+        """
+        Restore variables from file
+        :param file_path: file contain saved parameters.
+                          i.e. some parameters are obtained during training,
+                          not in trial config, e.g. scaler fit params)
+        :param config: the trial config
+        :return:
+        """
+        pass

pyzoo/zoo/automl/impute/impute.py

@@ -0,0 +1,44 @@
+#
+# Copyright 2018 Analytics Zoo Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+from zoo.automl.common.util import save_config
+from zoo.automl.impute.abstract import BaseImputation
+from zoo.zouwu.preprocessing.impute.LastFill import LastFill
+
+
+class LastFillImpute(BaseImputation):
+    """
+    LastFill imputation
+    """
+    def __init__(self):
+        self.imputer = LastFill()
+
+    def impute(self, input_df):
+        assert self.imputer is not None
+        df = self.imputer.impute(input_df)
+        return df
+
+    def restore(self, **config):
+        self.imputer = LastFill()
+
+
+class FillZeroImpute(BaseImputation):
+    """
+    FillZero imputation
+    """
+    def impute(self, input_df):
+        input_df.fillna(0)
+        return input_df

pyzoo/zoo/automl/regression/xgbregressor_predictor.py

@@ -78,7 +78,7 @@ def fit(self,
             mc=False,
             resources_per_trial={"cpu": 2},
             distributed=False,
-            hdfs_url=None
+            hdfs_url=None,
             ):
         """
         Trains the model for time sequence prediction.
@@ -121,7 +121,8 @@ def fit(self,
             recipe=recipe,
             mc=mc,
             resources_per_trial=resources_per_trial,
-            remote_dir=remote_dir)
+            remote_dir=remote_dir,
+        )
         return self.pipeline
 
     def evaluate(self,
@@ -220,10 +221,10 @@ def _hp_search(self,
                    recipe,
                    mc,
                    resources_per_trial,
-                   remote_dir):
+                   remote_dir,):
         def model_create_func():
             model = XGBoostRegressor(config)
-            model.set_params(n_jobs=resources_per_trial)
+            model.set_params(n_jobs=resources_per_trial.get("cpu", 1))
             return model
         model = model_create_func()
         ft = IdentityTransformer(self.feature_cols, self.target_col)
@@ -257,7 +258,7 @@ def model_create_func():
                          metric=metric,
                          metric_mode=metric_mode,
                          mc=mc,
-                         num_samples=num_samples)
+                         num_samples=num_samples,)
         # searcher.test_run()
         analysis = searcher.run()
 

pyzoo/zoo/automl/search/RayTuneSearchEngine.py

@@ -20,6 +20,7 @@
 
 from zoo.automl.search.abstract import *
 from zoo.automl.common.util import *
+from zoo.automl.impute.impute import *
 from ray.tune import Trainable
 import ray.tune.track
 from ray.tune.suggest.bayesopt import BayesOptSearch
@@ -102,7 +103,8 @@ def compile(self,
                                                    metric=metric,
                                                    metric_mode=metric_mode,
                                                    mc=mc,
-                                                   remote_dir=self.remote_dir)
+                                                   remote_dir=self.remote_dir
+                                                   )
         # self.trainable_class = self._prepare_trainable_class(input_df,
         #                                                      feature_transformers,
         #                                                      # model,
@@ -205,7 +207,7 @@ def _prepare_train_func(input_df,
                             metric_mode,
                             validation_df=None,
                             mc=False,
-                            remote_dir=None
+                            remote_dir=None,
                             ):
         """
         Prepare the train function for ray tune
@@ -244,9 +246,18 @@ def train_func(config):
             #
             trial_model = model_create_func
 
+            imputer = None
+            if "imputation" in config:
+                if config["imputation"] == "LastFillImpute":
+                    imputer = LastFillImpute()
+                elif config["imputation"] == "FillZeroImpute":
+                    imputer = FillZeroImpute()
+
             # handling input
             global_input_df = ray.get(input_df_id)
             trial_input_df = deepcopy(global_input_df)
+            if imputer:
+                trial_input_df = imputer.impute(trial_input_df)
             config = convert_bayes_configs(config).copy()
             # print("config is ", config)
             (x_train, y_train) = trial_ft.fit_transform(trial_input_df, **config)