fix: adjust CatBoost depth dynamically based on loss_function (#39)

* sensible depth for catboost

adjusting depth to sensible values to fit VRAM for GPU mode.

* fix: adjust CatBoost depth dynamically based on loss_function

Dynamically adjust max_depth for CatBoost based on task_type and loss_function:
- GPU pairwise modes (YetiRank, PairLogitPairwise, QueryCrossEntropy): max_depth=8
- GPU other modes: max_depth=16
- CPU: max_depth=16

Per CatBoost documentation, pairwise loss functions on GPU are limited to depth 8.

Co-authored-by: jokedoke <13506976+jokedoke@users.noreply.github.com>
* chore: bump strategy version to 3.10.9

* chore: bump model version to 3.10.9

---------

Co-authored-by: jokedoke <jokedoke@gmail.com>
Co-authored-by: jokedoke <13506976+jokedoke@users.noreply.github.com>

diff --git a/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py b/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py
index 03249cc4803750ec14264b8a7cd65e7b36e0905a..c79c1fa600c615ce2918434eca33db619adf3017 100644 (file)
--- a/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py
+++ b/quickadapter/user_data/freqaimodels/QuickAdapterRegressorV3.py
@@ -87,7 +87,7 @@ class QuickAdapterRegressorV3(BaseRegressionModel):
     https://github.com/sponsors/robcaulk
     """
 
-    version = "3.10.8"
+    version = "3.10.9"
 
     _TEST_SIZE: Final[float] = 0.1
 

diff --git a/quickadapter/user_data/strategies/QuickAdapterV3.py b/quickadapter/user_data/strategies/QuickAdapterV3.py
index 583c076f39fb45dc9733d07bea6c26ad1a33e1bb..62dc85977d810bdd9cf31891ae1fbf079968639d 100644 (file)
--- a/quickadapter/user_data/strategies/QuickAdapterV3.py
+++ b/quickadapter/user_data/strategies/QuickAdapterV3.py
@@ -110,7 +110,7 @@ class QuickAdapterV3(IStrategy):
     _PLOT_EXTREMA_MIN_EPS: Final[float] = 0.01
 
     def version(self) -> str:
-        return "3.10.8"
+        return "3.10.9"
 
     timeframe = "5m"
     timeframe_minutes = timeframe_to_minutes(timeframe)

diff --git a/quickadapter/user_data/strategies/Utils.py b/quickadapter/user_data/strategies/Utils.py
index 13f87a152373e00328ed366530282bd821fd8327..95120b0d0d88730704225114ce82fee0f314e03d 100644 (file)
--- a/quickadapter/user_data/strategies/Utils.py
+++ b/quickadapter/user_data/strategies/Utils.py
@@ -1632,6 +1632,12 @@ _CATBOOST_GPU_RSM_LOSS_FUNCTIONS: Final[tuple[str, ...]] = (
     "PairLogitPairwise",
 )
 
+_CATBOOST_GPU_PAIRWISE_LOSS_FUNCTIONS: Final[tuple[str, ...]] = (
+    "YetiRank",
+    "PairLogitPairwise",
+    "QueryCrossEntropy",
+)
+
 
 def get_ngboost_dist(dist_name: str) -> type:
     from ngboost.distns import Exponential, Laplace, LogNormal, Normal, T
@@ -2443,13 +2449,25 @@ def get_optuna_study_model_parameters(
     elif regressor == REGRESSORS[4]:  # "catboost"
         # Parameter order: boosting -> tree structure -> regularization -> sampling
         task_type = model_training_parameters.get("task_type", "CPU")
+        loss_function = model_training_parameters.get("loss_function", "RMSE")
+
+        if (
+            task_type == "GPU"
+            and loss_function in _CATBOOST_GPU_PAIRWISE_LOSS_FUNCTIONS
+        ):
+            max_depth = 8
+        elif task_type == "GPU":
+            max_depth = 16
+        else:  # CPU
+            max_depth = 16
+
         if task_type == "GPU":
             default_ranges: dict[str, tuple[float, float]] = {
                 # Boosting/Training
                 "iterations": (100, 2000),
                 "learning_rate": (0.001, 0.3),
                 # Tree structure
-                "depth": (4, 12),
+                "depth": (4, max_depth),
                 "min_data_in_leaf": (1, 20),
                 "border_count": (128, 255),
                 "max_ctr_complexity": (2, 6),
@@ -2470,7 +2488,7 @@ def get_optuna_study_model_parameters(
                 "iterations": (100, 2000),
                 "learning_rate": (0.001, 0.3),
                 # Tree structure
-                "depth": (4, 10),
+                "depth": (4, max_depth),
                 "min_data_in_leaf": (1, 20),
                 # Regularization
                 "l2_leaf_reg": (1, 10),
@@ -2550,7 +2568,6 @@ def get_optuna_study_model_parameters(
             ),
         }
 
-        loss_function = model_training_parameters.get("loss_function", "RMSE")
         if task_type == "CPU" or loss_function in _CATBOOST_GPU_RSM_LOSS_FUNCTIONS:
             params["rsm"] = trial.suggest_float(
                 "rsm",