perf(qav3): fine tune labelling window optimisation

Signed-off-by: Jérôme Benoit <jerome.benoit@sap.com>

diff --git a/quickadapter/user_data/freqaimodels/LightGBMRegressorQuickAdapterV35.py b/quickadapter/user_data/freqaimodels/LightGBMRegressorQuickAdapterV35.py
index f30ae26feeddbc7bd75c83910c9c31a83d5215a5..86725c902949f77c8c87e583e2affc9528ad132e 100644 (file)
--- a/quickadapter/user_data/freqaimodels/LightGBMRegressorQuickAdapterV35.py
+++ b/quickadapter/user_data/freqaimodels/LightGBMRegressorQuickAdapterV35.py
@@ -71,9 +71,9 @@ class LightGBMRegressorQuickAdapterV35(BaseRegressionModel):
             )
             self.freqai_info["feature_parameters"][pair] = {}
             self.freqai_info["feature_parameters"][pair]["label_period_candles"] = (
-                self.__optuna_period_params[pair].get(
-                    "label_period_candles", self.ft_params["label_period_candles"]
-                )
+                self.__optuna_period_params[
+                    pair
+                ].get("label_period_candles", self.ft_params["label_period_candles"])
             )
 
     def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
@@ -230,9 +230,9 @@ class LightGBMRegressorQuickAdapterV35(BaseRegressionModel):
         dk.data["extra_returns_per_train"]["DI_cutoff"] = cutoff
 
         dk.data["extra_returns_per_train"]["label_period_candles"] = (
-            self.__optuna_period_params.get(pair, {}).get(
-                "label_period_candles", self.ft_params["label_period_candles"]
-            )
+            self.__optuna_period_params.get(
+                pair, {}
+            ).get("label_period_candles", self.ft_params["label_period_candles"])
         )
         dk.data["extra_returns_per_train"]["hp_rmse"] = self.__optuna_hp_rmse.get(
             pair, -1
@@ -494,7 +494,7 @@ def log_sum_exp_min_max_pred(
     pred_df: pd.DataFrame, fit_live_predictions_candles: int, label_period_candles: int
 ) -> tuple[float, float]:
     label_period_frequency: int = int(
-        fit_live_predictions_candles / label_period_candles
+        fit_live_predictions_candles / (label_period_candles * 2)
     )
     extrema = pred_df.tail(label_period_candles * label_period_frequency)[
         EXTREMA_COLUMN
@@ -516,7 +516,7 @@ def mean_min_max_pred(
     )
 
     label_period_frequency: int = int(
-        fit_live_predictions_candles / label_period_candles
+        fit_live_predictions_candles / (label_period_candles * 2)
     )
     min_pred = pred_df_sorted.iloc[-label_period_frequency:].mean()
     max_pred = pred_df_sorted.iloc[:label_period_frequency].mean()
@@ -533,7 +533,7 @@ def median_min_max_pred(
     )
 
     label_period_frequency: int = int(
-        fit_live_predictions_candles / label_period_candles
+        fit_live_predictions_candles / (label_period_candles * 2)
     )
     min_pred = pred_df_sorted.iloc[-label_period_frequency:].median()
     max_pred = pred_df_sorted.iloc[:label_period_frequency].median()
@@ -592,8 +592,8 @@ def period_objective(
     )
     y_pred = model.predict(X_test)
 
-    min_label_period_candles = int(fit_live_predictions_candles / 6)
-    max_label_period_candles = int(fit_live_predictions_candles / 2)
+    min_label_period_candles = int(fit_live_predictions_candles / 20)
+    max_label_period_candles = int(fit_live_predictions_candles / 4)
     label_period_candles = trial.suggest_int(
         "label_period_candles",
         min_label_period_candles,

diff --git a/quickadapter/user_data/freqaimodels/XGBoostRegressorQuickAdapterV35.py b/quickadapter/user_data/freqaimodels/XGBoostRegressorQuickAdapterV35.py
index 052095c15b8bbbf5d127d53075188ace32c55673..46546463c542645ab88a79eaf339d3e15628af24 100644 (file)
--- a/quickadapter/user_data/freqaimodels/XGBoostRegressorQuickAdapterV35.py
+++ b/quickadapter/user_data/freqaimodels/XGBoostRegressorQuickAdapterV35.py
@@ -71,9 +71,9 @@ class XGBoostRegressorQuickAdapterV35(BaseRegressionModel):
             )
             self.freqai_info["feature_parameters"][pair] = {}
             self.freqai_info["feature_parameters"][pair]["label_period_candles"] = (
-                self.__optuna_period_params[pair].get(
-                    "label_period_candles", self.ft_params["label_period_candles"]
-                )
+                self.__optuna_period_params[
+                    pair
+                ].get("label_period_candles", self.ft_params["label_period_candles"])
             )
 
     def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
@@ -231,9 +231,9 @@ class XGBoostRegressorQuickAdapterV35(BaseRegressionModel):
         dk.data["extra_returns_per_train"]["DI_cutoff"] = cutoff
 
         dk.data["extra_returns_per_train"]["label_period_candles"] = (
-            self.__optuna_period_params.get(pair, {}).get(
-                "label_period_candles", self.ft_params["label_period_candles"]
-            )
+            self.__optuna_period_params.get(
+                pair, {}
+            ).get("label_period_candles", self.ft_params["label_period_candles"])
         )
         dk.data["extra_returns_per_train"]["hp_rmse"] = self.__optuna_hp_rmse.get(
             pair, -1
@@ -495,7 +495,7 @@ def log_sum_exp_min_max_pred(
     pred_df: pd.DataFrame, fit_live_predictions_candles: int, label_period_candles: int
 ) -> tuple[float, float]:
     label_period_frequency: int = int(
-        fit_live_predictions_candles / label_period_candles
+        fit_live_predictions_candles / (label_period_candles * 2)
     )
     extrema = pred_df.tail(label_period_candles * label_period_frequency)[
         EXTREMA_COLUMN
@@ -517,7 +517,7 @@ def mean_min_max_pred(
     )
 
     label_period_frequency: int = int(
-        fit_live_predictions_candles / label_period_candles
+        fit_live_predictions_candles / (label_period_candles * 2)
     )
     min_pred = pred_df_sorted.iloc[-label_period_frequency:].mean()
     max_pred = pred_df_sorted.iloc[:label_period_frequency].mean()
@@ -534,7 +534,7 @@ def median_min_max_pred(
     )
 
     label_period_frequency: int = int(
-        fit_live_predictions_candles / label_period_candles
+        fit_live_predictions_candles / (label_period_candles * 2)
     )
     min_pred = pred_df_sorted.iloc[-label_period_frequency:].median()
     max_pred = pred_df_sorted.iloc[:label_period_frequency].median()
@@ -596,8 +596,8 @@ def period_objective(
     )
     y_pred = model.predict(X_test)
 
-    min_label_period_candles = int(fit_live_predictions_candles / 6)
-    max_label_period_candles = int(fit_live_predictions_candles / 2)
+    min_label_period_candles = int(fit_live_predictions_candles / 20)
+    max_label_period_candles = int(fit_live_predictions_candles / 4)
     label_period_candles = trial.suggest_int(
         "label_period_candles",
         min_label_period_candles,

diff --git a/quickadapter/user_data/strategies/QuickAdapterV3.py b/quickadapter/user_data/strategies/QuickAdapterV3.py
index 89b309aca9746ffad7fa02bf71ce12c6db430fd2..9b36469a9b1c428d717dc6b958bcc4e3043735e0 100644 (file)
--- a/quickadapter/user_data/strategies/QuickAdapterV3.py
+++ b/quickadapter/user_data/strategies/QuickAdapterV3.py
@@ -10,7 +10,7 @@ from freqtrade.strategy.interface import IStrategy
 from technical.pivots_points import pivots_points
 from freqtrade.exchange import timeframe_to_prev_date
 from freqtrade.persistence import Trade
-from scipy.signal import find_peaks
+from scipy.signal import argrelmin, argrelmax
 import numpy as np
 import pandas_ta as pta
 
@@ -248,19 +248,31 @@ class QuickAdapterV3(IStrategy):
                 self.freqai_info["feature_parameters"]["label_period_candles"],
             )
         )
-        min_peaks, _ = find_peaks(
-            -dataframe["low"].values,
-            distance=label_period_candles,
+        min_peaks = argrelmin(
+            dataframe["low"].values,
+            order=label_period_candles,
         )
-        max_peaks, _ = find_peaks(
+        max_peaks = argrelmax(
             dataframe["high"].values,
-            distance=label_period_candles,
+            order=label_period_candles,
         )
+        # min_peaks, _ = find_peaks(
+        #     -dataframe["low"].values,
+        #     distance=label_period_candles,
+        # )
+        # max_peaks, _ = find_peaks(
+        #     dataframe["high"].values,
+        #     distance=label_period_candles,
+        # )
         dataframe[EXTREMA_COLUMN] = 0
-        for mp in min_peaks:
+        for mp in min_peaks[0]:
             dataframe.at[mp, EXTREMA_COLUMN] = -1
-        for mp in max_peaks:
+        for mp in max_peaks[0]:
             dataframe.at[mp, EXTREMA_COLUMN] = 1
+        # for mp in min_peaks:
+        #     dataframe.at[mp, EXTREMA_COLUMN] = -1
+        # for mp in max_peaks:
+        #     dataframe.at[mp, EXTREMA_COLUMN] = 1
         dataframe["minima"] = np.where(dataframe[EXTREMA_COLUMN] == -1, -1, 0)
         dataframe["maxima"] = np.where(dataframe[EXTREMA_COLUMN] == 1, 1, 0)
         dataframe[EXTREMA_COLUMN] = (