Merge pull request #7367 from freqtrade/add-continual-learning

add continual learning to catboost and friends

docs/freqai.md

@@ -98,6 +98,7 @@ Mandatory parameters are marked as **Required**, which means that they are requi
 | `expiration_hours` | Avoid making predictions if a model is more than `expiration_hours` old. <br> Defaults set to 0, which means models never expire. <br> **Datatype:** Positive integer.
 | `fit_live_predictions_candles` | Number of historical candles to use for computing target (label) statistics from prediction data, instead of from the training data set. <br> **Datatype:** Positive integer.
 | `follow_mode` | If true, this instance of FreqAI will look for models associated with `identifier` and load those for inferencing. A `follower` will **not** train new models. <br> **Datatype:** Boolean. Default: `False`.
+| `continual_learning` | If true, FreqAI will start training new models from the final state of the most recently trained model. <br> **Datatype:** Boolean. Default: `False`.
 |  |  **Feature parameters**
 | `feature_parameters` | A dictionary containing the parameters used to engineer the feature set. Details and examples are shown [here](#feature-engineering). <br> **Datatype:** Dictionary.
 | `include_timeframes` | A list of timeframes that all indicators in `populate_any_indicators` will be created for. The list is added as features to the base asset feature set. <br> **Datatype:** List of timeframes (strings).

freqtrade/freqai/base_models/BaseClassifierModel.py

@@ -21,12 +21,12 @@ class BaseClassifierModel(IFreqaiModel):
     """
 
     def train(
-        self, unfiltered_dataframe: DataFrame, pair: str, dk: FreqaiDataKitchen
+        self, unfiltered_df: DataFrame, pair: str, dk: FreqaiDataKitchen, **kwargs
     ) -> Any:
         """
         Filter the training data and train a model to it. Train makes heavy use of the datakitchen
         for storing, saving, loading, and analyzing the data.
-        :param unfiltered_dataframe: Full dataframe for the current training period
+        :param unfiltered_df: Full dataframe for the current training period
         :param metadata: pair metadata from strategy.
         :return:
         :model: Trained model which can be used to inference (self.predict)
@@ -36,14 +36,14 @@ class BaseClassifierModel(IFreqaiModel):
 
         # filter the features requested by user in the configuration file and elegantly handle NaNs
         features_filtered, labels_filtered = dk.filter_features(
-            unfiltered_dataframe,
+            unfiltered_df,
             dk.training_features_list,
             dk.label_list,
             training_filter=True,
         )
 
-        start_date = unfiltered_dataframe["date"].iloc[0].strftime("%Y-%m-%d")
-        end_date = unfiltered_dataframe["date"].iloc[-1].strftime("%Y-%m-%d")
+        start_date = unfiltered_df["date"].iloc[0].strftime("%Y-%m-%d")
+        end_date = unfiltered_df["date"].iloc[-1].strftime("%Y-%m-%d")
         logger.info(f"-------------------- Training on data from {start_date} to "
                     f"{end_date}--------------------")
         # split data into train/test data.
@@ -61,32 +61,32 @@ class BaseClassifierModel(IFreqaiModel):
         )
         logger.info(f'Training model on {len(data_dictionary["train_features"])} data points')
 
-        model = self.fit(data_dictionary)
+        model = self.fit(data_dictionary, dk)
 
         logger.info(f"--------------------done training {pair}--------------------")
 
         return model
 
     def predict(
-        self, unfiltered_dataframe: DataFrame, dk: FreqaiDataKitchen, first: bool = False
+        self, unfiltered_df: DataFrame, dk: FreqaiDataKitchen, **kwargs
     ) -> Tuple[DataFrame, npt.NDArray[np.int_]]:
         """
         Filter the prediction features data and predict with it.
-        :param: unfiltered_dataframe: Full dataframe for the current backtest period.
+        :param: unfiltered_df: Full dataframe for the current backtest period.
         :return:
         :pred_df: dataframe containing the predictions
         :do_predict: np.array of 1s and 0s to indicate places where freqai needed to remove
         data (NaNs) or felt uncertain about data (PCA and DI index)
         """
 
-        dk.find_features(unfiltered_dataframe)
-        filtered_dataframe, _ = dk.filter_features(
-            unfiltered_dataframe, dk.training_features_list, training_filter=False
+        dk.find_features(unfiltered_df)
+        filtered_df, _ = dk.filter_features(
+            unfiltered_df, dk.training_features_list, training_filter=False
         )
-        filtered_dataframe = dk.normalize_data_from_metadata(filtered_dataframe)
-        dk.data_dictionary["prediction_features"] = filtered_dataframe
+        filtered_df = dk.normalize_data_from_metadata(filtered_df)
+        dk.data_dictionary["prediction_features"] = filtered_df
 
-        self.data_cleaning_predict(dk, filtered_dataframe)
+        self.data_cleaning_predict(dk, filtered_df)
 
         predictions = self.model.predict(dk.data_dictionary["prediction_features"])
         pred_df = DataFrame(predictions, columns=dk.label_list)

freqtrade/freqai/base_models/BaseRegressionModel.py

@@ -20,12 +20,12 @@ class BaseRegressionModel(IFreqaiModel):
     """
 
     def train(
-        self, unfiltered_dataframe: DataFrame, pair: str, dk: FreqaiDataKitchen
+        self, unfiltered_df: DataFrame, pair: str, dk: FreqaiDataKitchen, **kwargs
     ) -> Any:
         """
         Filter the training data and train a model to it. Train makes heavy use of the datakitchen
         for storing, saving, loading, and analyzing the data.
-        :param unfiltered_dataframe: Full dataframe for the current training period
+        :param unfiltered_df: Full dataframe for the current training period
         :param metadata: pair metadata from strategy.
         :return:
         :model: Trained model which can be used to inference (self.predict)
@@ -35,14 +35,14 @@ class BaseRegressionModel(IFreqaiModel):
 
         # filter the features requested by user in the configuration file and elegantly handle NaNs
         features_filtered, labels_filtered = dk.filter_features(
-            unfiltered_dataframe,
+            unfiltered_df,
             dk.training_features_list,
             dk.label_list,
             training_filter=True,
         )
 
-        start_date = unfiltered_dataframe["date"].iloc[0].strftime("%Y-%m-%d")
-        end_date = unfiltered_dataframe["date"].iloc[-1].strftime("%Y-%m-%d")
+        start_date = unfiltered_df["date"].iloc[0].strftime("%Y-%m-%d")
+        end_date = unfiltered_df["date"].iloc[-1].strftime("%Y-%m-%d")
         logger.info(f"-------------------- Training on data from {start_date} to "
                     f"{end_date}--------------------")
         # split data into train/test data.
@@ -60,33 +60,33 @@ class BaseRegressionModel(IFreqaiModel):
         )
         logger.info(f'Training model on {len(data_dictionary["train_features"])} data points')
 
-        model = self.fit(data_dictionary)
+        model = self.fit(data_dictionary, dk)
 
         logger.info(f"--------------------done training {pair}--------------------")
 
         return model
 
     def predict(
-        self, unfiltered_dataframe: DataFrame, dk: FreqaiDataKitchen, first: bool = False
+        self, unfiltered_df: DataFrame, dk: FreqaiDataKitchen, **kwargs
     ) -> Tuple[DataFrame, npt.NDArray[np.int_]]:
         """
         Filter the prediction features data and predict with it.
-        :param: unfiltered_dataframe: Full dataframe for the current backtest period.
+        :param: unfiltered_df: Full dataframe for the current backtest period.
         :return:
         :pred_df: dataframe containing the predictions
         :do_predict: np.array of 1s and 0s to indicate places where freqai needed to remove
         data (NaNs) or felt uncertain about data (PCA and DI index)
         """
 
-        dk.find_features(unfiltered_dataframe)
-        filtered_dataframe, _ = dk.filter_features(
-            unfiltered_dataframe, dk.training_features_list, training_filter=False
+        dk.find_features(unfiltered_df)
+        filtered_df, _ = dk.filter_features(
+            unfiltered_df, dk.training_features_list, training_filter=False
         )
-        filtered_dataframe = dk.normalize_data_from_metadata(filtered_dataframe)
-        dk.data_dictionary["prediction_features"] = filtered_dataframe
+        filtered_df = dk.normalize_data_from_metadata(filtered_df)
+        dk.data_dictionary["prediction_features"] = filtered_df
 
         # optional additional data cleaning/analysis
-        self.data_cleaning_predict(dk, filtered_dataframe)
+        self.data_cleaning_predict(dk, filtered_df)
 
         predictions = self.model.predict(dk.data_dictionary["prediction_features"])
         pred_df = DataFrame(predictions, columns=dk.label_list)

freqtrade/freqai/base_models/BaseTensorFlowModel.py

@@ -17,12 +17,12 @@ class BaseTensorFlowModel(IFreqaiModel):
     """
 
     def train(
-        self, unfiltered_dataframe: DataFrame, pair: str, dk: FreqaiDataKitchen
+        self, unfiltered_df: DataFrame, pair: str, dk: FreqaiDataKitchen, **kwargs
     ) -> Any:
         """
         Filter the training data and train a model to it. Train makes heavy use of the datakitchen
         for storing, saving, loading, and analyzing the data.
-        :param unfiltered_dataframe: Full dataframe for the current training period
+        :param unfiltered_df: Full dataframe for the current training period
         :param metadata: pair metadata from strategy.
         :return:
         :model: Trained model which can be used to inference (self.predict)
@@ -32,14 +32,14 @@ class BaseTensorFlowModel(IFreqaiModel):
 
         # filter the features requested by user in the configuration file and elegantly handle NaNs
         features_filtered, labels_filtered = dk.filter_features(
-            unfiltered_dataframe,
+            unfiltered_df,
             dk.training_features_list,
             dk.label_list,
             training_filter=True,
         )
 
-        start_date = unfiltered_dataframe["date"].iloc[0].strftime("%Y-%m-%d")
-        end_date = unfiltered_dataframe["date"].iloc[-1].strftime("%Y-%m-%d")
+        start_date = unfiltered_df["date"].iloc[0].strftime("%Y-%m-%d")
+        end_date = unfiltered_df["date"].iloc[-1].strftime("%Y-%m-%d")
         logger.info(f"-------------------- Training on data from {start_date} to "
                     f"{end_date}--------------------")
         # split data into train/test data.
@@ -57,7 +57,7 @@ class BaseTensorFlowModel(IFreqaiModel):
         )
         logger.info(f'Training model on {len(data_dictionary["train_features"])} data points')
 
-        model = self.fit(data_dictionary)
+        model = self.fit(data_dictionary, dk)
 
         logger.info(f"--------------------done training {pair}--------------------")
 

freqtrade/freqai/base_models/FreqaiMultiOutputRegressor.py

@@ -0,0 +1,75 @@
+
+from joblib import Parallel
+from sklearn.multioutput import MultiOutputRegressor, _fit_estimator
+from sklearn.utils.fixes import delayed
+from sklearn.utils.validation import has_fit_parameter
+
+
+class FreqaiMultiOutputRegressor(MultiOutputRegressor):
+
+    def fit(self, X, y, sample_weight=None, fit_params=None):
+        """Fit the model to data, separately for each output variable.
+        Parameters
+        ----------
+        X : {array-like, sparse matrix} of shape (n_samples, n_features)
+            The input data.
+        y : {array-like, sparse matrix} of shape (n_samples, n_outputs)
+            Multi-output targets. An indicator matrix turns on multilabel
+            estimation.
+        sample_weight : array-like of shape (n_samples,), default=None
+            Sample weights. If `None`, then samples are equally weighted.
+            Only supported if the underlying regressor supports sample
+            weights.
+        fit_params : A list of dicts for the fit_params
+            Parameters passed to the ``estimator.fit`` method of each step.
+            Each dict may contain same or different values (e.g. different
+            eval_sets or init_models)
+            .. versionadded:: 0.23
+        Returns
+        -------
+        self : object
+            Returns a fitted instance.
+        """
+
+        if not hasattr(self.estimator, "fit"):
+            raise ValueError("The base estimator should implement a fit method")
+
+        y = self._validate_data(X="no_validation", y=y, multi_output=True)
+
+        # if is_classifier(self):
+        #     check_classification_targets(y)
+
+        if y.ndim == 1:
+            raise ValueError(
+                "y must have at least two dimensions for "
+                "multi-output regression but has only one."
+            )
+
+        if sample_weight is not None and not has_fit_parameter(
+            self.estimator, "sample_weight"
+        ):
+            raise ValueError("Underlying estimator does not support sample weights.")
+
+        # fit_params_validated = _check_fit_params(X, fit_params)
+
+        if not fit_params:
+            fit_params = [None] * y.shape[1]
+
+        # if not init_models:
+        #     init_models = [None] * y.shape[1]
+
+        self.estimators_ = Parallel(n_jobs=self.n_jobs)(
+            delayed(_fit_estimator)(
+                self.estimator, X, y[:, i], sample_weight, **fit_params[i]
+                # init_model=init_models[i], eval_set=eval_sets[i],
+                # **fit_params_validated
+            )
+            for i in range(y.shape[1])
+        )
+
+        if hasattr(self.estimators_[0], "n_features_in_"):
+            self.n_features_in_ = self.estimators_[0].n_features_in_
+        if hasattr(self.estimators_[0], "feature_names_in_"):
+            self.feature_names_in_ = self.estimators_[0].feature_names_in_
+
+        return

freqtrade/freqai/freqai_interface.py

@@ -88,6 +88,7 @@ class IFreqaiModel(ABC):
         self.begin_time: float = 0
         self.begin_time_train: float = 0
         self.base_tf_seconds = timeframe_to_seconds(self.config['timeframe'])
+        self.continual_learning = self.freqai_info.get('continual_learning', False)
 
         self._threads: List[threading.Thread] = []
         self._stop_event = threading.Event()
@@ -676,21 +677,30 @@ class IFreqaiModel(ABC):
                 self.train_time = 0
         return
 
+    def get_init_model(self, pair: str) -> Any:
+        if pair not in self.dd.model_dictionary or not self.continual_learning:
+            init_model = None
+        else:
+            init_model = self.dd.model_dictionary[pair]
+
+        return init_model
+
     # Following methods which are overridden by user made prediction models.
     # See freqai/prediction_models/CatboostPredictionModel.py for an example.
 
     @abstractmethod
-    def train(self, unfiltered_dataframe: DataFrame, pair: str, dk: FreqaiDataKitchen) -> Any:
+    def train(self, unfiltered_df: DataFrame, pair: str,
+              dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         Filter the training data and train a model to it. Train makes heavy use of the datahandler
         for storing, saving, loading, and analyzing the data.
-        :param unfiltered_dataframe: Full dataframe for the current training period
+        :param unfiltered_df: Full dataframe for the current training period
         :param metadata: pair metadata from strategy.
         :return: Trained model which can be used to inference (self.predict)
         """
 
     @abstractmethod
-    def fit(self, data_dictionary: Dict[str, Any]) -> Any:
+    def fit(self, data_dictionary: Dict[str, Any], dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         Most regressors use the same function names and arguments e.g. user
         can drop in LGBMRegressor in place of CatBoostRegressor and all data
@@ -703,11 +713,11 @@ class IFreqaiModel(ABC):
 
     @abstractmethod
     def predict(
-        self, dataframe: DataFrame, dk: FreqaiDataKitchen, first: bool = True
+        self, unfiltered_df: DataFrame, dk: FreqaiDataKitchen, **kwargs
     ) -> Tuple[DataFrame, NDArray[np.int_]]:
         """
         Filter the prediction features data and predict with it.
-        :param unfiltered_dataframe: Full dataframe for the current backtest period.
+        :param unfiltered_df: Full dataframe for the current backtest period.
         :param dk: FreqaiDataKitchen = Data management/analysis tool associated to present pair only
         :param first: boolean = whether this is the first prediction or not.
         :return:

freqtrade/freqai/prediction_models/CatboostClassifier.py

@@ -3,7 +3,8 @@ from typing import Any, Dict
 
 from catboost import CatBoostClassifier, Pool
 
-from freqtrade.freqai.prediction_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.base_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 
 logger = logging.getLogger(__name__)
@@ -16,7 +17,7 @@ class CatboostClassifier(BaseClassifierModel):
     has its own DataHandler where data is held, saved, loaded, and managed.
     """
 
-    def fit(self, data_dictionary: Dict) -> Any:
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :params:
@@ -36,6 +37,8 @@ class CatboostClassifier(BaseClassifierModel):
             **self.model_training_parameters,
         )
 
-        cbr.fit(train_data)
+        init_model = self.get_init_model(dk.pair)
+
+        cbr.fit(train_data, init_model=init_model)
 
         return cbr

freqtrade/freqai/prediction_models/CatboostRegressor.py

@@ -1,10 +1,10 @@
-import gc
 import logging
 from typing import Any, Dict
 
 from catboost import CatBoostRegressor, Pool
 
-from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 
 logger = logging.getLogger(__name__)
@@ -17,7 +17,7 @@ class CatboostRegressor(BaseRegressionModel):
     has its own DataHandler where data is held, saved, loaded, and managed.
     """
 
-    def fit(self, data_dictionary: Dict) -> Any:
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :param data_dictionary: the dictionary constructed by DataHandler to hold
@@ -38,16 +38,13 @@ class CatboostRegressor(BaseRegressionModel):
                 weight=data_dictionary["test_weights"],
             )
 
+        init_model = self.get_init_model(dk.pair)
+
         model = CatBoostRegressor(
             allow_writing_files=False,
             **self.model_training_parameters,
         )
 
-        model.fit(X=train_data, eval_set=test_data)
-
-        # some evidence that catboost pools have memory leaks:
-        # https://github.com/catboost/catboost/issues/1835
-        del train_data, test_data
-        gc.collect()
+        model.fit(X=train_data, eval_set=test_data, init_model=init_model)
 
         return model

freqtrade/freqai/prediction_models/CatboostRegressorMultiTarget.py

@@ -1,10 +1,11 @@
 import logging
 from typing import Any, Dict
 
-from catboost import CatBoostRegressor  # , Pool
-from sklearn.multioutput import MultiOutputRegressor
+from catboost import CatBoostRegressor, Pool
 
-from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.FreqaiMultiOutputRegressor import FreqaiMultiOutputRegressor
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 
 logger = logging.getLogger(__name__)
@@ -17,7 +18,7 @@ class CatboostRegressorMultiTarget(BaseRegressionModel):
     has its own DataHandler where data is held, saved, loaded, and managed.
     """
 
-    def fit(self, data_dictionary: Dict) -> Any:
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :param data_dictionary: the dictionary constructed by DataHandler to hold
@@ -31,14 +32,34 @@ class CatboostRegressorMultiTarget(BaseRegressionModel):
 
         X = data_dictionary["train_features"]
         y = data_dictionary["train_labels"]
-        eval_set = (data_dictionary["test_features"], data_dictionary["test_labels"])
+
         sample_weight = data_dictionary["train_weights"]
 
-        model = MultiOutputRegressor(estimator=cbr)
-        model.fit(X=X, y=y, sample_weight=sample_weight)  # , eval_set=eval_set)
+        eval_sets = [None] * y.shape[1]
 
         if self.freqai_info.get('data_split_parameters', {}).get('test_size', 0.1) != 0:
-            train_score = model.score(X, y)
-            test_score = model.score(*eval_set)
-            logger.info(f"Train score {train_score}, Test score {test_score}")
+            eval_sets = [None] * data_dictionary['test_labels'].shape[1]
+
+            for i in range(data_dictionary['test_labels'].shape[1]):
+                eval_sets[i] = Pool(
+                    data=data_dictionary["test_features"],
+                    label=data_dictionary["test_labels"].iloc[:, i],
+                    weight=data_dictionary["test_weights"],
+                )
+
+        init_model = self.get_init_model(dk.pair)
+
+        if init_model:
+            init_models = init_model.estimators_
+        else:
+            init_models = [None] * y.shape[1]
+
+        fit_params = []
+        for i in range(len(eval_sets)):
+            fit_params.append(
+                {'eval_set': eval_sets[i],  'init_model': init_models[i]})
+
+        model = FreqaiMultiOutputRegressor(estimator=cbr)
+        model.fit(X=X, y=y, sample_weight=sample_weight, fit_params=fit_params)
+
         return model

freqtrade/freqai/prediction_models/LightGBMClassifier.py

@@ -3,7 +3,8 @@ from typing import Any, Dict
 
 from lightgbm import LGBMClassifier
 
-from freqtrade.freqai.prediction_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.base_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 
 logger = logging.getLogger(__name__)
@@ -16,7 +17,7 @@ class LightGBMClassifier(BaseClassifierModel):
     has its own DataHandler where data is held, saved, loaded, and managed.
     """
 
-    def fit(self, data_dictionary: Dict) -> Any:
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :params:
@@ -35,9 +36,11 @@ class LightGBMClassifier(BaseClassifierModel):
         y = data_dictionary["train_labels"].to_numpy()[:, 0]
         train_weights = data_dictionary["train_weights"]
 
+        init_model = self.get_init_model(dk.pair)
+
         model = LGBMClassifier(**self.model_training_parameters)
 
         model.fit(X=X, y=y, eval_set=eval_set, sample_weight=train_weights,
-                  eval_sample_weight=[test_weights])
+                  eval_sample_weight=[test_weights], init_model=init_model)
 
         return model

freqtrade/freqai/prediction_models/LightGBMRegressor.py

@@ -3,7 +3,8 @@ from typing import Any, Dict
 
 from lightgbm import LGBMRegressor
 
-from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 
 logger = logging.getLogger(__name__)
@@ -16,7 +17,7 @@ class LightGBMRegressor(BaseRegressionModel):
     has its own DataHandler where data is held, saved, loaded, and managed.
     """
 
-    def fit(self, data_dictionary: Dict) -> Any:
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         Most regressors use the same function names and arguments e.g. user
         can drop in LGBMRegressor in place of CatBoostRegressor and all data
@@ -35,9 +36,11 @@ class LightGBMRegressor(BaseRegressionModel):
         y = data_dictionary["train_labels"]
         train_weights = data_dictionary["train_weights"]
 
+        init_model = self.get_init_model(dk.pair)
+
         model = LGBMRegressor(**self.model_training_parameters)
 
         model.fit(X=X, y=y, eval_set=eval_set, sample_weight=train_weights,
-                  eval_sample_weight=[eval_weights])
+                  eval_sample_weight=[eval_weights], init_model=init_model)
 
         return model

freqtrade/freqai/prediction_models/LightGBMRegressorMultiTarget.py

@@ -2,9 +2,10 @@ import logging
 from typing import Any, Dict
 
 from lightgbm import LGBMRegressor
-from sklearn.multioutput import MultiOutputRegressor
 
-from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.FreqaiMultiOutputRegressor import FreqaiMultiOutputRegressor
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 
 logger = logging.getLogger(__name__)
@@ -17,7 +18,7 @@ class LightGBMRegressorMultiTarget(BaseRegressionModel):
     has its own DataHandler where data is held, saved, loaded, and managed.
     """
 
-    def fit(self, data_dictionary: Dict) -> Any:
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :param data_dictionary: the dictionary constructed by DataHandler to hold
@@ -28,12 +29,36 @@ class LightGBMRegressorMultiTarget(BaseRegressionModel):
 
         X = data_dictionary["train_features"]
         y = data_dictionary["train_labels"]
-        eval_set = (data_dictionary["test_features"], data_dictionary["test_labels"])
         sample_weight = data_dictionary["train_weights"]
 
-        model = MultiOutputRegressor(estimator=lgb)
-        model.fit(X=X, y=y, sample_weight=sample_weight)  # , eval_set=eval_set)
-        train_score = model.score(X, y)
-        test_score = model.score(*eval_set)
-        logger.info(f"Train score {train_score}, Test score {test_score}")
+        eval_weights = None
+        eval_sets = [None] * y.shape[1]
+
+        if self.freqai_info.get('data_split_parameters', {}).get('test_size', 0.1) != 0:
+            eval_weights = [data_dictionary["test_weights"]]
+            eval_sets = [(None, None)] * data_dictionary['test_labels'].shape[1]  # type: ignore
+            for i in range(data_dictionary['test_labels'].shape[1]):
+                eval_sets[i] = (  # type: ignore
+                    data_dictionary["test_features"],
+                    data_dictionary["test_labels"].iloc[:, i]
+                )
+
+        init_model = self.get_init_model(dk.pair)
+        if init_model:
+            init_models = init_model.estimators_
+        else:
+            init_models = [None] * y.shape[1]
+
+        fit_params = []
+        for i in range(len(eval_sets)):
+            fit_params.append(
+                {'eval_set': eval_sets[i], 'eval_sample_weight': eval_weights,
+                 'init_model': init_models[i]})
+
+        model = FreqaiMultiOutputRegressor(estimator=lgb)
+        model.fit(X=X, y=y, sample_weight=sample_weight, fit_params=fit_params)
+
+        # model = FreqaiMultiOutputRegressor(estimator=lgb)
+        # model.fit(X=X, y=y, sample_weight=sample_weight, init_models=init_models,
+        #           eval_sets=eval_sets, eval_sample_weight=eval_weights)
         return model

freqtrade/freqai/prediction_models/XGBoostRegressor.py

@@ -0,0 +1,45 @@
+import logging
+from typing import Any, Dict
+
+from xgboost import XGBRegressor
+
+from freqtrade.freqai.base_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
+
+
+logger = logging.getLogger(__name__)
+
+
+class XGBoostRegressor(BaseRegressionModel):
+    """
+    User created prediction model. The class needs to override three necessary
+    functions, predict(), train(), fit(). The class inherits ModelHandler which
+    has its own DataHandler where data is held, saved, loaded, and managed.
+    """
+
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
+        """
+        User sets up the training and test data to fit their desired model here
+        :param data_dictionary: the dictionary constructed by DataHandler to hold
+                                all the training and test data/labels.
+        """
+
+        X = data_dictionary["train_features"]
+        y = data_dictionary["train_labels"]
+
+        if self.freqai_info.get("data_split_parameters", {}).get("test_size", 0.1) == 0:
+            eval_set = None
+        else:
+            eval_set = [(data_dictionary["test_features"], data_dictionary["test_labels"])]
+            eval_weights = [data_dictionary['test_weights']]
+
+        sample_weight = data_dictionary["train_weights"]
+
+        xgb_model = self.get_init_model(dk.pair)
+
+        model = XGBRegressor(**self.model_training_parameters)
+
+        model.fit(X=X, y=y, sample_weight=sample_weight, eval_set=eval_set,
+                  sample_weight_eval_set=eval_weights, xgb_model=xgb_model)
+
+        return model

freqtrade/freqai/prediction_models/XGBoostRegressorMultiTarget.py

@@ -0,0 +1,60 @@
+import logging
+from typing import Any, Dict
+
+from xgboost import XGBRegressor
+
+from freqtrade.freqai.base_models.BaseRegressionModel import BaseRegressionModel
+from freqtrade.freqai.base_models.FreqaiMultiOutputRegressor import FreqaiMultiOutputRegressor
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
+
+
+logger = logging.getLogger(__name__)
+
+
+class XGBoostRegressorMultiTarget(BaseRegressionModel):
+    """
+    User created prediction model. The class needs to override three necessary
+    functions, predict(), train(), fit(). The class inherits ModelHandler which
+    has its own DataHandler where data is held, saved, loaded, and managed.
+    """
+
+    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
+        """
+        User sets up the training and test data to fit their desired model here
+        :param data_dictionary: the dictionary constructed by DataHandler to hold
+                                all the training and test data/labels.
+        """
+
+        xgb = XGBRegressor(**self.model_training_parameters)
+
+        X = data_dictionary["train_features"]
+        y = data_dictionary["train_labels"]
+        sample_weight = data_dictionary["train_weights"]
+
+        eval_weights = None
+        eval_sets = [None] * y.shape[1]
+
+        if self.freqai_info.get('data_split_parameters', {}).get('test_size', 0.1) != 0:
+            eval_weights = [data_dictionary["test_weights"]]
+            for i in range(data_dictionary['test_labels'].shape[1]):
+                eval_sets[i] = [(  # type: ignore
+                    data_dictionary["test_features"],
+                    data_dictionary["test_labels"].iloc[:, i]
+                )]
+
+        init_model = self.get_init_model(dk.pair)
+        if init_model:
+            init_models = init_model.estimators_
+        else:
+            init_models = [None] * y.shape[1]
+
+        fit_params = []
+        for i in range(len(eval_sets)):
+            fit_params.append(
+                {'eval_set': eval_sets[i], 'sample_weight_eval_set': eval_weights,
+                 'xgb_model': init_models[i]})
+
+        model = FreqaiMultiOutputRegressor(estimator=xgb)
+        model.fit(X=X, y=y, sample_weight=sample_weight, fit_params=fit_params)
+
+        return model

requirements-freqai.txt

@@ -6,3 +6,4 @@ scikit-learn==1.1.2
 joblib==1.1.0
 catboost==1.0.6; platform_machine != 'aarch64'
 lightgbm==3.3.2
+xgboost==1.6.2

tests/freqai/test_freqai_interface.py

@@ -174,6 +174,69 @@ def test_extract_data_and_train_model_LightGBMClassifier(mocker, freqai_conf):
     shutil.rmtree(Path(freqai.dk.full_path))
 
 
+def test_extract_data_and_train_model_XGBoostRegressor(mocker, freqai_conf):
+    freqai_conf.update({"timerange": "20180110-20180130"})
+    freqai_conf.update({"freqaimodel": "XGBoostRegressor"})
+    freqai_conf.update({"strategy": "freqai_test_strat"})
+
+    strategy = get_patched_freqai_strategy(mocker, freqai_conf)
+    exchange = get_patched_exchange(mocker, freqai_conf)
+    strategy.dp = DataProvider(freqai_conf, exchange)
+    strategy.freqai_info = freqai_conf.get("freqai", {})
+    freqai = strategy.freqai
+    freqai.live = True
+    freqai.dk = FreqaiDataKitchen(freqai_conf)
+    timerange = TimeRange.parse_timerange("20180110-20180130")
+    freqai.dd.load_all_pair_histories(timerange, freqai.dk)
+
+    freqai.dd.pair_dict = MagicMock()
+
+    data_load_timerange = TimeRange.parse_timerange("20180110-20180130")
+    new_timerange = TimeRange.parse_timerange("20180120-20180130")
+
+    freqai.extract_data_and_train_model(
+        new_timerange, "ADA/BTC", strategy, freqai.dk, data_load_timerange)
+
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_model.joblib").is_file()
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_metadata.json").is_file()
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_trained_df.pkl").is_file()
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_svm_model.joblib").is_file()
+
+    shutil.rmtree(Path(freqai.dk.full_path))
+
+
+def test_extract_data_and_train_model_XGBoostRegressorMultiModel(mocker, freqai_conf):
+    freqai_conf.update({"timerange": "20180110-20180130"})
+    freqai_conf.update({"freqaimodel": "XGBoostRegressorMultiTarget"})
+    freqai_conf.update({"strategy": "freqai_test_multimodel_strat"})
+    strategy = get_patched_freqai_strategy(mocker, freqai_conf)
+    exchange = get_patched_exchange(mocker, freqai_conf)
+    strategy.dp = DataProvider(freqai_conf, exchange)
+    strategy.freqai_info = freqai_conf.get("freqai", {})
+    freqai = strategy.freqai
+    freqai.live = True
+    freqai.dk = FreqaiDataKitchen(freqai_conf)
+    timerange = TimeRange.parse_timerange("20180110-20180130")
+    freqai.dd.load_all_pair_histories(timerange, freqai.dk)
+
+    freqai.dd.pair_dict = MagicMock()
+
+    data_load_timerange = TimeRange.parse_timerange("20180110-20180130")
+    new_timerange = TimeRange.parse_timerange("20180120-20180130")
+
+    freqai.extract_data_and_train_model(
+        new_timerange, "ADA/BTC", strategy, freqai.dk, data_load_timerange)
+
+    assert len(freqai.dk.label_list) == 2
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_model.joblib").is_file()
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_metadata.json").is_file()
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_trained_df.pkl").is_file()
+    assert Path(freqai.dk.data_path / f"{freqai.dk.model_filename}_svm_model.joblib").is_file()
+    assert len(freqai.dk.data['training_features_list']) == 26
+
+    shutil.rmtree(Path(freqai.dk.full_path))
+
+
 def test_start_backtesting(mocker, freqai_conf):
     freqai_conf.update({"timerange": "20180120-20180130"})
     freqai_conf.get("freqai", {}).update({"save_backtest_models": True})