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/freqai_interface.py

@@ -86,6 +86,7 @@ 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()
@@ -674,7 +675,7 @@ class IFreqaiModel(ABC):
         """
 
     @abstractmethod
-    def fit(self, data_dictionary: Dict[str, Any]) -> Any:
+    def fit(self, data_dictionary: Dict[str, Any], dk: FreqaiDataKitchen) -> Any:
         """
         Most regressors use the same function names and arguments e.g. user
         can drop in LGBMRegressor in place of CatBoostRegressor and all data

freqtrade/freqai/prediction_models/BaseClassifierModel.py

@@ -61,7 +61,7 @@ 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}--------------------")
 

freqtrade/freqai/prediction_models/BaseRegressionModel.py

@@ -60,7 +60,7 @@ 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}--------------------")
 

freqtrade/freqai/prediction_models/BaseTensorFlowModel.py

@@ -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/prediction_models/CatboostClassifier.py

@@ -2,7 +2,7 @@ import logging
 from typing import Any, Dict
 
 from catboost import CatBoostClassifier, Pool
-
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 from freqtrade.freqai.prediction_models.BaseClassifierModel import BaseClassifierModel
 
 
@@ -16,7 +16,7 @@ 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) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :params:
@@ -36,6 +36,11 @@ class CatboostClassifier(BaseClassifierModel):
             **self.model_training_parameters,
         )
 
-        cbr.fit(train_data)
+        if dk.pair not in self.dd.model_dictionary or not self.continual_learning:
+            init_model = None
+        else:
+            init_model = self.dd.model_dictionary[dk.pair]
+
+        cbr.fit(train_data, init_model=init_model)
 
         return cbr

freqtrade/freqai/prediction_models/CatboostRegressor.py

@@ -3,6 +3,7 @@ import logging
 from typing import Any, Dict
 
 from catboost import CatBoostRegressor, Pool
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
 
@@ -17,7 +18,7 @@ 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) -> 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 +39,16 @@ class CatboostRegressor(BaseRegressionModel):
                 weight=data_dictionary["test_weights"],
             )
 
+        if dk.pair not in self.dd.model_dictionary or not self.continual_learning:
+            init_model = None
+        else:
+            init_model = self.dd.model_dictionary[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

@@ -3,7 +3,7 @@ from typing import Any, Dict
 
 from catboost import CatBoostRegressor  # , Pool
 from sklearn.multioutput import MultiOutputRegressor
-
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
 
 
@@ -17,7 +17,7 @@ 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) -> 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
@@ -34,6 +34,9 @@ class CatboostRegressorMultiTarget(BaseRegressionModel):
         eval_set = (data_dictionary["test_features"], data_dictionary["test_labels"])
         sample_weight = data_dictionary["train_weights"]
 
+        if self.continual_learning:
+            logger.warning('Continual learning not supported for MultiTarget models')
+
         model = MultiOutputRegressor(estimator=cbr)
         model.fit(X=X, y=y, sample_weight=sample_weight)  # , eval_set=eval_set)
 

freqtrade/freqai/prediction_models/LightGBMClassifier.py

@@ -4,7 +4,7 @@ from typing import Any, Dict
 from lightgbm import LGBMClassifier
 
 from freqtrade.freqai.prediction_models.BaseClassifierModel import BaseClassifierModel
-
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 logger = logging.getLogger(__name__)
 
@@ -16,7 +16,7 @@ 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) -> Any:
         """
         User sets up the training and test data to fit their desired model here
         :params:
@@ -35,9 +35,14 @@ class LightGBMClassifier(BaseClassifierModel):
         y = data_dictionary["train_labels"].to_numpy()[:, 0]
         train_weights = data_dictionary["train_weights"]
 
+        if dk.pair not in self.dd.model_dictionary or not self.continual_learning:
+            init_model = None
+        else:
+            init_model = self.dd.model_dictionary[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

@@ -4,7 +4,7 @@ from typing import Any, Dict
 from lightgbm import LGBMRegressor
 
 from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
-
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 logger = logging.getLogger(__name__)
 
@@ -16,7 +16,7 @@ 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) -> 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 +35,14 @@ class LightGBMRegressor(BaseRegressionModel):
         y = data_dictionary["train_labels"]
         train_weights = data_dictionary["train_weights"]
 
+        if dk.pair not in self.dd.model_dictionary or not self.continual_learning:
+            init_model = None
+        else:
+            init_model = self.dd.model_dictionary[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

@@ -5,7 +5,7 @@ from lightgbm import LGBMRegressor
 from sklearn.multioutput import MultiOutputRegressor
 
 from freqtrade.freqai.prediction_models.BaseRegressionModel import BaseRegressionModel
-
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 
 logger = logging.getLogger(__name__)
 
@@ -17,7 +17,7 @@ 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) -> 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
@@ -31,6 +31,9 @@ class LightGBMRegressorMultiTarget(BaseRegressionModel):
         eval_set = (data_dictionary["test_features"], data_dictionary["test_labels"])
         sample_weight = data_dictionary["train_weights"]
 
+        if self.continual_learning:
+            logger.warning('Continual learning not supported for MultiTarget models')
+
         model = MultiOutputRegressor(estimator=lgb)
         model.fit(X=X, y=y, sample_weight=sample_weight)  # , eval_set=eval_set)
         train_score = model.score(X, y)