Merge branch 'develop' into feat/externalsignals

freqtrade/data/dataprovider.py

@@ -313,7 +313,9 @@ class DataProvider:
         Clear pair dataframe cache.
         """
         self.__cached_pairs = {}
-        self.__cached_pairs_backtesting = {}
+        # Don't reset backtesting pairs -
+        # otherwise they're reloaded each time during hyperopt due to with analyze_per_epoch
+        # self.__cached_pairs_backtesting = {}
         self.__slice_index = 0
 
     # Exchange functions

freqtrade/freqai/data_drawer.py

@@ -355,7 +355,7 @@ class FreqaiDataDrawer:
         for dir in model_folders:
             result = pattern.match(str(dir.name))
             if result is None:
-                break
+                continue
             coin = result.group(1)
             timestamp = result.group(2)
 

freqtrade/freqai/prediction_models/XGBoostClassifier.py

@@ -0,0 +1,85 @@
+import logging
+from typing import Any, Dict, Tuple
+
+import numpy as np
+import numpy.typing as npt
+import pandas as pd
+from pandas import DataFrame
+from pandas.api.types import is_integer_dtype
+from sklearn.preprocessing import LabelEncoder
+from xgboost import XGBClassifier
+
+from freqtrade.freqai.base_models.BaseClassifierModel import BaseClassifierModel
+from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
+
+
+logger = logging.getLogger(__name__)
+
+
+class XGBoostClassifier(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
+        :params:
+        :data_dictionary: the dictionary constructed by DataHandler to hold
+        all the training and test data/labels.
+        """
+
+        X = data_dictionary["train_features"].to_numpy()
+        y = data_dictionary["train_labels"].to_numpy()[:, 0]
+
+        le = LabelEncoder()
+        if not is_integer_dtype(y):
+            y = pd.Series(le.fit_transform(y), dtype="int64")
+
+        if self.freqai_info.get('data_split_parameters', {}).get('test_size', 0.1) == 0:
+            eval_set = None
+        else:
+            test_features = data_dictionary["test_features"].to_numpy()
+            test_labels = data_dictionary["test_labels"].to_numpy()[:, 0]
+
+            if not is_integer_dtype(test_labels):
+                test_labels = pd.Series(le.transform(test_labels), dtype="int64")
+
+            eval_set = [(test_features, test_labels)]
+
+        train_weights = data_dictionary["train_weights"]
+
+        init_model = self.get_init_model(dk.pair)
+
+        model = XGBClassifier(**self.model_training_parameters)
+
+        model.fit(X=X, y=y, eval_set=eval_set, sample_weight=train_weights,
+                  xgb_model=init_model)
+
+        return model
+
+    def predict(
+        self, unfiltered_df: DataFrame, dk: FreqaiDataKitchen, **kwargs
+    ) -> Tuple[DataFrame, npt.NDArray[np.int_]]:
+        """
+        Filter the prediction features data and predict with it.
+        :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)
+        """
+
+        (pred_df, dk.do_predict) = super().predict(unfiltered_df, dk, **kwargs)
+
+        le = LabelEncoder()
+        label = dk.label_list[0]
+        labels_before = list(dk.data['labels_std'].keys())
+        labels_after = le.fit_transform(labels_before).tolist()
+        pred_df[label] = le.inverse_transform(pred_df[label])
+        pred_df = pred_df.rename(
+            columns={labels_after[i]: labels_before[i] for i in range(len(labels_before))})
+
+        return (pred_df, dk.do_predict)

freqtrade/optimize/hyperopt.py

@@ -580,11 +580,23 @@ class Hyperopt:
                     max_value=self.total_epochs, redirect_stdout=False, redirect_stderr=False,
                     widgets=widgets
                 ) as pbar:
-                    EVALS = ceil(self.total_epochs / jobs)
-                    for i in range(EVALS):
+                    start = 0
+
+                    if self.analyze_per_epoch:
+                        # First analysis not in parallel mode when using --analyze-per-epoch.
+                        # This allows dataprovider to load it's informative cache.
+                        asked, is_random = self.get_asked_points(n_points=1)
+                        f_val0 = self.generate_optimizer(asked[0])
+                        self.opt.tell(asked, [f_val0['loss']])
+                        self.evaluate_result(f_val0, 1, is_random[0])
+                        pbar.update(1)
+                        start += 1
+
+                    evals = ceil((self.total_epochs - start) / jobs)
+                    for i in range(evals):
                         # Correct the number of epochs to be processed for the last
                         # iteration (should not exceed self.total_epochs in total)
-                        n_rest = (i + 1) * jobs - self.total_epochs
+                        n_rest = (i + 1) * jobs - (self.total_epochs - start)
                         current_jobs = jobs - n_rest if n_rest > 0 else jobs
 
                         asked, is_random = self.get_asked_points(n_points=current_jobs)
@@ -594,7 +606,7 @@ class Hyperopt:
                         # Calculate progressbar outputs
                         for j, val in enumerate(f_val):
                             # Use human-friendly indexes here (starting from 1)
-                            current = i * jobs + j + 1
+                            current = i * jobs + j + 1 + start
 
                             self.evaluate_result(val, current, is_random[j])