fix(qav3): ensure pivot labeling optimization can run only once per new

candle

Signed-off-by: Jérôme Benoit <jerome.benoit@piment-noir.org>

diff --git a/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py b/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py
index 32df9fd1720ba824829d356eb3047b770aef284b..9f3511211e3b670edf581dd4cbabfe0aefab5194 100644 (file)
--- a/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py
+++ b/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py
@@ -48,7 +48,7 @@ class QuickAdapterRegressorV3(BaseRegressionModel):
     https://github.com/sponsors/robcaulk
     """
 
-    version = "3.7.82"
+    version = "3.7.83"
 
     @cached_property
     def _optuna_config(self) -> dict:
@@ -97,6 +97,7 @@ class QuickAdapterRegressorV3(BaseRegressionModel):
         self._optuna_hp_params: dict[str, dict] = {}
         self._optuna_train_params: dict[str, dict] = {}
         self._optuna_label_params: dict[str, dict] = {}
+        self.init_optuna_label_candle_pool()
         self._optuna_label_candles: dict[str, int] = {}
         self._optuna_label_candle: dict[str, int] = {}
         for pair in self.pairs:
@@ -126,7 +127,7 @@ class QuickAdapterRegressorV3(BaseRegressionModel):
                 }
             )
             self._optuna_label_candles[pair] = 0
-            self._optuna_label_candle[pair] = self.get_optuna_label_candle()
+            self.set_optuna_label_candle(pair)
 
         logger.info(
             f"Initialized {self.__class__.__name__} {self.freqai_info.get('regressor', 'xgboost')} regressor model version {self.version}"
@@ -183,14 +184,36 @@ class QuickAdapterRegressorV3(BaseRegressionModel):
         else:
             raise ValueError(f"Invalid namespace: {namespace}")
 
-    def get_optuna_label_candle(self) -> int:
+    def get_optuna_label_all_candles(self) -> list[int]:
+        n_pairs = len(self.pairs)
         label_frequency_candles = max(
-            2, int(self.ft_params.get("label_frequency_candles", 12))
+            2, n_pairs, int(self.ft_params.get("label_frequency_candles", 12))
         )
-        random_offset = random.randint(
-            -label_frequency_candles // 2, label_frequency_candles // 2
+        min_offset = -int(label_frequency_candles / 2)
+        max_offset = int(label_frequency_candles / 2)
+        return [
+            max(1, label_frequency_candles + offset)
+            for offset in range(min_offset, max_offset + 1)
+        ]
+
+    def init_optuna_label_candle_pool(self) -> None:
+        self._optuna_label_candle_pool = self.get_optuna_label_all_candles()
+        random.shuffle(self._optuna_label_candle_pool)
+        if len(self._optuna_label_candle_pool) == 0:
+            raise RuntimeError("Failed to initialize optuna label candle pool")
+
+    def set_optuna_label_candle(self, pair: str) -> None:
+        if len(self._optuna_label_candle_pool) == 0:
+            self.init_optuna_label_candle_pool()
+        self._optuna_label_candle[pair] = self._optuna_label_candle_pool.pop()
+        optuna_label_available_candles = (
+            set(self.get_optuna_label_all_candles())
+            - set(self._optuna_label_candle_pool)
+            - set(self._optuna_label_candle.values())
         )
-        return max(1, label_frequency_candles + random_offset)
+        if len(optuna_label_available_candles) > 0:
+            self._optuna_label_candle_pool.extend(optuna_label_available_candles)
+            random.shuffle(self._optuna_label_candle_pool)
 
     def fit(self, data_dictionary: dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
         """
@@ -304,7 +327,7 @@ class QuickAdapterRegressorV3(BaseRegressionModel):
                 )
             finally:
                 self._optuna_label_candles[pair] = 0
-                self._optuna_label_candle[pair] = self.get_optuna_label_candle()
+                self.set_optuna_label_candle(pair)
         else:
             logger.info(
                 f"Optuna {pair} {namespace} callback throttled, still {self._optuna_label_candle[pair] - self._optuna_label_candles[pair]} candles to go"
@@ -1029,7 +1052,7 @@ def train_objective(
     candles_step: int,
     model_training_parameters: dict,
 ) -> float:
-    min_test_window: int = fit_live_predictions_candles * 2
+    min_test_window: int = fit_live_predictions_candles * 4
     if len(X_test) < min_test_window:
         logger.warning(f"Insufficient test data: {len(X_test)} < {min_test_window}")
         return np.inf

diff --git a/quickadapter/user_data/strategies/QuickAdapterV3.py b/quickadapter/user_data/strategies/QuickAdapterV3.py
index 102b4840f2f3e6441d5fb67cb49d365b50d3f6b3..2d77e8394a8d5137a1326cae2db5b5b9a044f0b8 100644 (file)
--- a/quickadapter/user_data/strategies/QuickAdapterV3.py
+++ b/quickadapter/user_data/strategies/QuickAdapterV3.py
@@ -61,7 +61,7 @@ class QuickAdapterV3(IStrategy):
     INTERFACE_VERSION = 3
 
     def version(self) -> str:
-        return "3.3.86"
+        return "3.3.87"
 
     timeframe = "5m"