add xgbregressor

Author: frederikhoengaard

python/src/lazylearn/lazylearn.py

@@ -50,9 +50,13 @@ def run_autopilot(self):
 
         :return:
         """
-        sb = StrategyBuilder(task=self.task, dataset=self.dataset, target=self.target, random_state=self.random_state)
+        sb = StrategyBuilder(
+            task=self.task,
+            dataset=self.dataset,
+            target=self.target,
+            random_state=self.random_state,
+        )
         self.leaderboard = sorted([model for model in sb.models], key=lambda x: x.score)
 
     def get_leaderboard(self):
         return [(item.name, item.score) for item in self.leaderboard]
-

python/src/lazylearn/regression/models/xgboost/xgb.py

@@ -0,0 +1,39 @@
+from models.models import Dataset
+from pipeline.pipeline import RegressionPipeline
+from preprocessing.encoding.encoders import OrdinalConverter
+from regression.models.xgboost.xgb_regressor_steps.regressor_step import XGBRegressorStep
+
+
+class XGBRegressionRunner:
+    def __init__(self, target, dataset, random_state=None):
+        self.name = "XGBRegressor"
+        self.target = target
+        self.dataset: Dataset = dataset
+        self.random_state = random_state
+        self.pipeline = RegressionPipeline()
+        self.pipeline.target = target
+
+        self.pipeline.train_features_df = self.dataset.partitions[
+            "train"
+        ].copy()  # noqa
+        self.pipeline.train_targets = self.dataset.partitions["train"][target]
+        self.pipeline.holdout_features_df = self.dataset.partitions[
+            "test"
+        ].copy()  # noqa
+        self.pipeline.holdout_targets = self.dataset.partitions["test"][target]
+
+    def fit(self):
+        # preprocess numeric vars
+        cat_vars = self.dataset.type_collections["categorical"]
+        num_vars = self.dataset.type_collections["numeric"]
+        self.pipeline.feature_list.extend(num_vars)
+
+        self.pipeline.add(OrdinalConverter(cat_vars=cat_vars))
+
+        self.pipeline.add(XGBRegressorStep(random_state=self.random_state))
+
+        self.pipeline.fit()
+
+    def predict(self, features):
+        self.pipeline.tmp_test = features
+        return self.pipeline.predict()

python/src/lazylearn/regression/models/xgboost/xgb_regressor_steps/__init__.py

@@ -0,0 +1,23 @@
+from pipeline.pipeline import PipelineStep, RegressionPipeline
+from xgboost import XGBRegressor
+
+
+class XGBRegressorStep(PipelineStep):
+    def __init__(self, random_state=None):
+        self.regressor = XGBRegressor(n_estimators=1000, random_state=random_state)
+
+    def fit(self, pipeline: RegressionPipeline):
+        pipeline.feature_list = [
+            item for item in pipeline.feature_list if item != pipeline.target
+        ]
+        print("Fitting XGBRegressor")
+        self.regressor.fit(
+            X=pipeline.train_features_df[pipeline.feature_list],
+            y=pipeline.train_targets,
+        )  # noqa
+        print("XGBRegressor fitted!")
+
+    def predict(self, pipeline: RegressionPipeline):
+        pipeline.tmp_pred = self.regressor.predict(
+            X=pipeline.tmp_test[pipeline.feature_list]
+        )

python/src/lazylearn/regression/models/xgboost/xgb_regressor_steps/regressor_step.py

@@ -1,7 +1,8 @@
-from sklearn.metrics import mean_absolute_error
-
 from models.models import Model
 from regression.models.randomforest.randomforest import RandomForestRegressionRunner
+from regression.models.xgboost.xgb import XGBRegressionRunner
+from sklearn.metrics import mean_absolute_error
+
 
 class StrategyBuilder:
     def __init__(self, task, dataset, target, random_state=None):
@@ -16,14 +17,23 @@ def __init__(self, task, dataset, target, random_state=None):
         self.start()
 
     def build(self):
+        # TODO: Make sure pipelines are not overwriting data for each other
         if self.task == "regression":
+            self.strategies.append(
+                XGBRegressionRunner(
+                    target=self.target,
+                    dataset=self.dataset,
+                    random_state=self.random_state  # noqa
+                )
+            )
             self.strategies.append(
                 RandomForestRegressionRunner(
                     target=self.target,
                     dataset=self.dataset,
                     random_state=self.random_state,  # noqa
                 )
             )
+
         else:
             raise ValueError("Unsupported task!")
 
@@ -38,5 +48,10 @@ def start(self):
                 strategy.pipeline.tmp_pred,
             )
 
-            self.models.append(Model(name=strategy.name, score=strategy.pipeline.holdout_score, pipeline=strategy))
-
+            self.models.append(
+                Model(
+                    name=strategy.name,
+                    score=strategy.pipeline.holdout_score,
+                    pipeline=strategy,
+                )
+            )