Merge pull request #7644 from markdregan/multi-target-classifier

Support for multi target multi-class classification (FreqaiMultiOutputRegressor for Classification)

freqtrade/freqai/base_models/FreqaiMultiOutputClassifier.py

@@ -0,0 +1,93 @@
+import numpy as np
+from joblib import Parallel
+from sklearn.base import is_classifier
+from sklearn.multioutput import MultiOutputClassifier, _fit_estimator
+from sklearn.utils.fixes import delayed
+from sklearn.utils.multiclass import check_classification_targets
+from sklearn.utils.validation import has_fit_parameter
+
+from freqtrade.exceptions import OperationalException
+
+
+class FreqaiMultiOutputClassifier(MultiOutputClassifier):
+
+    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 classifier 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.")
+
+        if not fit_params:
+            fit_params = [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]
+            )
+            for i in range(y.shape[1])
+        )
+
+        self.classes_ = []
+        for estimator in self.estimators_:
+            self.classes_.extend(estimator.classes_)
+        if len(set(self.classes_)) != len(self.classes_):
+            raise OperationalException(f"Class labels must be unique across targets: "
+                                       f"{self.classes_}")
+
+        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 self
+
+    def predict_proba(self, X):
+        """
+        Get predict_proba and stack arrays horizontally
+        """
+        results = np.hstack(super().predict_proba(X))
+        return np.squeeze(results)
+
+    def predict(self, X):
+        """
+        Get predict and squeeze into 2D array
+        """
+        results = super().predict(X)
+        return np.squeeze(results)

freqtrade/freqai/prediction_models/CatboostClassifierMultiTarget.py

@@ -0,0 +1,74 @@
+import logging
+import sys
+from pathlib import Path
+from typing import Any, Dict
+
+from catboost import CatBoostClassifier, Pool
+
+from freqtrade.freqai.base_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.base_models.FreqaiMultiOutputClassifier import FreqaiMultiOutputClassifier
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
+
+
+logger = logging.getLogger(__name__)
+
+
+class CatboostClassifierMultiTarget(BaseClassifierModel):
+    """
+    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.
+        """
+
+        cbc = CatBoostClassifier(
+            allow_writing_files=True,
+            loss_function='MultiClass',
+            train_dir=Path(dk.data_path),
+            **self.model_training_parameters,
+        )
+
+        X = data_dictionary["train_features"]
+        y = data_dictionary["train_labels"]
+
+        sample_weight = data_dictionary["train_weights"]
+
+        eval_sets = [None] * y.shape[1]
+
+        if self.freqai_info.get('data_split_parameters', {}).get('test_size', 0.1) != 0:
+            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],
+                'log_cout': sys.stdout, 'log_cerr': sys.stderr,
+            })
+
+        model = FreqaiMultiOutputClassifier(estimator=cbc)
+        thread_training = self.freqai_info.get('multitarget_parallel_training', False)
+        if thread_training:
+            model.n_jobs = y.shape[1]
+        model.fit(X=X, y=y, sample_weight=sample_weight, fit_params=fit_params)
+
+        return model

freqtrade/freqai/prediction_models/LightGBMClassifierMultiTarget.py

@@ -0,0 +1,64 @@
+import logging
+from typing import Any, Dict
+
+from lightgbm import LGBMClassifier
+
+from freqtrade.freqai.base_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.base_models.FreqaiMultiOutputClassifier import FreqaiMultiOutputClassifier
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
+
+
+logger = logging.getLogger(__name__)
+
+
+class LightGBMClassifierMultiTarget(BaseClassifierModel):
+    """
+    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.
+        """
+
+        lgb = LGBMClassifier(**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"]]
+            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 = FreqaiMultiOutputClassifier(estimator=lgb)
+        thread_training = self.freqai_info.get('multitarget_parallel_training', False)
+        if thread_training:
+            model.n_jobs = y.shape[1]
+        model.fit(X=X, y=y, sample_weight=sample_weight, fit_params=fit_params)
+
+        return model

tests/freqai/test_freqai_interface.py

@@ -77,17 +77,19 @@ def test_extract_data_and_train_model_Standard(mocker, freqai_conf, model, pca,
     shutil.rmtree(Path(freqai.dk.full_path))
 
 
-@pytest.mark.parametrize('model', [
+@pytest.mark.parametrize('model, strat', [
-    'LightGBMRegressorMultiTarget',
+    ('LightGBMRegressorMultiTarget', "freqai_test_multimodel_strat"),
-    'XGBoostRegressorMultiTarget',
+    ('XGBoostRegressorMultiTarget', "freqai_test_multimodel_strat"),
-    'CatboostRegressorMultiTarget',
+    ('CatboostRegressorMultiTarget', "freqai_test_multimodel_strat"),
+    ('LightGBMClassifierMultiTarget', "freqai_test_multimodel_classifier_strat"),
+    ('CatboostClassifierMultiTarget', "freqai_test_multimodel_classifier_strat")
     ])
-def test_extract_data_and_train_model_MultiTargets(mocker, freqai_conf, model):
+def test_extract_data_and_train_model_MultiTargets(mocker, freqai_conf, model, strat):
-    if is_arm() and model == 'CatboostRegressorMultiTarget':
+    if is_arm() and 'Catboost' in model:
         pytest.skip("CatBoost is not supported on ARM")
 
     freqai_conf.update({"timerange": "20180110-20180130"})
-    freqai_conf.update({"strategy": "freqai_test_multimodel_strat"})
+    freqai_conf.update({"strategy": strat})
     freqai_conf.update({"freqaimodel": model})
     strategy = get_patched_freqai_strategy(mocker, freqai_conf)
     exchange = get_patched_exchange(mocker, freqai_conf)

tests/rpc/test_rpc_apiserver.py

@@ -1460,6 +1460,7 @@ def test_api_strategies(botclient, tmpdir):
         'StrategyTestV3CustomEntryPrice',
         'StrategyTestV3Futures',
         'freqai_test_classifier',
+        'freqai_test_multimodel_classifier_strat',
         'freqai_test_multimodel_strat',
         'freqai_test_strat'
     ]}

tests/strategy/strats/freqai_test_multimodel_classifier_strat.py

@@ -0,0 +1,138 @@
+import logging
+from functools import reduce
+
+import numpy as np
+import pandas as pd
+import talib.abstract as ta
+from pandas import DataFrame
+
+from freqtrade.strategy import DecimalParameter, IntParameter, IStrategy, merge_informative_pair
+
+
+logger = logging.getLogger(__name__)
+
+
+class freqai_test_multimodel_classifier_strat(IStrategy):
+    """
+    Test strategy - used for testing freqAI multimodel functionalities.
+    DO not use in production.
+    """
+
+    minimal_roi = {"0": 0.1, "240": -1}
+
+    plot_config = {
+        "main_plot": {},
+        "subplots": {
+            "prediction": {"prediction": {"color": "blue"}},
+            "target_roi": {
+                "target_roi": {"color": "brown"},
+            },
+            "do_predict": {
+                "do_predict": {"color": "brown"},
+            },
+        },
+    }
+
+    process_only_new_candles = True
+    stoploss = -0.05
+    use_exit_signal = True
+    startup_candle_count: int = 300
+    can_short = False
+
+    linear_roi_offset = DecimalParameter(
+        0.00, 0.02, default=0.005, space="sell", optimize=False, load=True
+    )
+    max_roi_time_long = IntParameter(0, 800, default=400, space="sell", optimize=False, load=True)
+
+    def populate_any_indicators(
+        self, pair, df, tf, informative=None, set_generalized_indicators=False
+    ):
+
+        coin = pair.split('/')[0]
+
+        if informative is None:
+            informative = self.dp.get_pair_dataframe(pair, tf)
+
+        # first loop is automatically duplicating indicators for time periods
+        for t in self.freqai_info["feature_parameters"]["indicator_periods_candles"]:
+
+            t = int(t)
+            informative[f"%-{coin}rsi-period_{t}"] = ta.RSI(informative, timeperiod=t)
+            informative[f"%-{coin}mfi-period_{t}"] = ta.MFI(informative, timeperiod=t)
+            informative[f"%-{coin}adx-period_{t}"] = ta.ADX(informative, window=t)
+
+        informative[f"%-{coin}pct-change"] = informative["close"].pct_change()
+        informative[f"%-{coin}raw_volume"] = informative["volume"]
+        informative[f"%-{coin}raw_price"] = informative["close"]
+
+        indicators = [col for col in informative if col.startswith("%")]
+        # This loop duplicates and shifts all indicators to add a sense of recency to data
+        for n in range(self.freqai_info["feature_parameters"]["include_shifted_candles"] + 1):
+            if n == 0:
+                continue
+            informative_shift = informative[indicators].shift(n)
+            informative_shift = informative_shift.add_suffix("_shift-" + str(n))
+            informative = pd.concat((informative, informative_shift), axis=1)
+
+        df = merge_informative_pair(df, informative, self.config["timeframe"], tf, ffill=True)
+        skip_columns = [
+            (s + "_" + tf) for s in ["date", "open", "high", "low", "close", "volume"]
+        ]
+        df = df.drop(columns=skip_columns)
+
+        # Add generalized indicators here (because in live, it will call this
+        # function to populate indicators during training). Notice how we ensure not to
+        # add them multiple times
+        if set_generalized_indicators:
+            df["%-day_of_week"] = (df["date"].dt.dayofweek + 1) / 7
+            df["%-hour_of_day"] = (df["date"].dt.hour + 1) / 25
+
+            # user adds targets here by prepending them with &- (see convention below)
+            # If user wishes to use multiple targets, a multioutput prediction model
+            # needs to be used such as templates/CatboostPredictionMultiModel.py
+            df['&s-up_or_down'] = np.where(df["close"].shift(-50) >
+                                           df["close"], 'up', 'down')
+
+            df['&s-up_or_down2'] = np.where(df["close"].shift(-50) >
+                                            df["close"], 'up2', 'down2')
+
+        return df
+
+    def populate_indicators(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
+
+        self.freqai_info = self.config["freqai"]
+
+        dataframe = self.freqai.start(dataframe, metadata, self)
+
+        dataframe["target_roi"] = dataframe["&-s_close_mean"] + dataframe["&-s_close_std"] * 1.25
+        dataframe["sell_roi"] = dataframe["&-s_close_mean"] - dataframe["&-s_close_std"] * 1.25
+        return dataframe
+
+    def populate_entry_trend(self, df: DataFrame, metadata: dict) -> DataFrame:
+
+        enter_long_conditions = [df["do_predict"] == 1, df["&-s_close"] > df["target_roi"]]
+
+        if enter_long_conditions:
+            df.loc[
+                reduce(lambda x, y: x & y, enter_long_conditions), ["enter_long", "enter_tag"]
+            ] = (1, "long")
+
+        enter_short_conditions = [df["do_predict"] == 1, df["&-s_close"] < df["sell_roi"]]
+
+        if enter_short_conditions:
+            df.loc[
+                reduce(lambda x, y: x & y, enter_short_conditions), ["enter_short", "enter_tag"]
+            ] = (1, "short")
+
+        return df
+
+    def populate_exit_trend(self, df: DataFrame, metadata: dict) -> DataFrame:
+        exit_long_conditions = [df["do_predict"] == 1, df["&-s_close"] < df["sell_roi"] * 0.25]
+        if exit_long_conditions:
+            df.loc[reduce(lambda x, y: x & y, exit_long_conditions), "exit_long"] = 1
+
+        exit_short_conditions = [df["do_predict"] == 1, df["&-s_close"] > df["target_roi"] * 0.25]
+        if exit_short_conditions:
+            df.loc[reduce(lambda x, y: x & y, exit_short_conditions), "exit_short"] = 1
+
+        return df

tests/strategy/test_strategy_loading.py

@@ -34,7 +34,7 @@ def test_search_all_strategies_no_failed():
     directory = Path(__file__).parent / "strats"
     strategies = StrategyResolver._search_all_objects(directory, enum_failed=False)
     assert isinstance(strategies, list)
-    assert len(strategies) == 10
+    assert len(strategies) == 11
     assert isinstance(strategies[0], dict)
 
 
@@ -42,10 +42,10 @@ def test_search_all_strategies_with_failed():
     directory = Path(__file__).parent / "strats"
     strategies = StrategyResolver._search_all_objects(directory, enum_failed=True)
     assert isinstance(strategies, list)
-    assert len(strategies) == 11
+    assert len(strategies) == 12
     # with enum_failed=True search_all_objects() shall find 2 good strategies
     # and 1 which fails to load
-    assert len([x for x in strategies if x['class'] is not None]) == 10
+    assert len([x for x in strategies if x['class'] is not None]) == 11
 
     assert len([x for x in strategies if x['class'] is None]) == 1