autogluon.tabular 1.2.1b20250425__tar.gz → 1.2.1b20250426__tar.gz

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.tabular
-Version: 1.2.1b20250425
+Version: 1.2.1b20250426
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon/tabular/models/catboost/catboost_model.py

@@ -95,17 +95,26 @@ class CatBoostModel(AbstractModel):
             Scales roughly by 5080*num_features*2^depth bytes
             For 10000 features and 6 depth, the histogram would be 3.2 GB.
         """
+        if hyperparameters is None:
+            hyperparameters = {}
         num_classes = num_classes if num_classes else 1  # self.num_classes could be None after initialization if it's a regression problem
         data_mem_usage = get_approximate_df_mem_usage(X).sum()
         data_mem_usage_bytes = data_mem_usage * 5 + data_mem_usage / 4 * num_classes  # TODO: Extremely crude approximation, can be vastly improved
 
         border_count = hyperparameters.get("border_count", 254)
         depth = hyperparameters.get("depth", 6)
+
+        # if depth < 7, treat it as 1 step larger for histogram size estimate
+        #  this fixes cases where otherwise histogram size appears to be off by around a factor of 2 for depth=6
+        histogram_effective_depth = max(min(depth+1, 7), depth)
+
         # Formula based on manual testing, aligns with LightGBM histogram sizes
-        histogram_mem_usage_bytes = 20 * math.pow(2, depth) * len(X.columns) * border_count
+        histogram_mem_usage_bytes = 24 * math.pow(2, histogram_effective_depth) * len(X.columns) * border_count
         histogram_mem_usage_bytes *= 1.2  # Add a 20% buffer
 
-        approx_mem_size_req = data_mem_usage_bytes + histogram_mem_usage_bytes
+        baseline_memory_bytes = 4e8  # 400 MB baseline memory
+
+        approx_mem_size_req = data_mem_usage_bytes + histogram_mem_usage_bytes + baseline_memory_bytes
         return approx_mem_size_req
 
     # TODO: Use Pool in preprocess, optimize bagging to do Pool.split() to avoid re-computing pool for each fold! Requires stateful + y

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon/tabular/models/lgb/lgb_model.py

@@ -3,7 +3,6 @@ from __future__ import annotations
 import gc
 import logging
 import os
-import random
 import re
 import time
 import warnings
@@ -99,6 +98,8 @@ class LGBModel(AbstractModel):
             Scales roughly by 5100*num_features*num_leaves bytes
             For 10000 features and 128 num_leaves, the histogram would be 6.5 GB.
         """
+        if hyperparameters is None:
+            hyperparameters = {}
         num_classes = num_classes if num_classes else 1  # num_classes could be None after initialization if it's a regression problem
         data_mem_usage = get_approximate_df_mem_usage(X).sum()
         data_mem_usage_bytes = data_mem_usage * 5 + data_mem_usage / 4 * num_classes  # TODO: Extremely crude approximation, can be vastly improved
@@ -275,8 +276,6 @@ class LGBModel(AbstractModel):
             train_params["params"]["quantile_levels"] = self.quantile_levels
         if seed_val is not None:
             train_params["params"]["seed"] = seed_val
-            random.seed(seed_val)
-            np.random.seed(seed_val)
 
         # Train LightGBM model:
         # Note that self.model contains a <class 'lightgbm.basic.Booster'> not a LightBGMClassifier or LightGBMRegressor object

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon/tabular/models/rf/rf_model.py

@@ -147,6 +147,8 @@ class RFModel(AbstractModel):
         num_classes: int = 1,
         **kwargs,
     ) -> int:
+        if hyperparameters is None:
+            hyperparameters = {}
         n_estimators_final = hyperparameters.get("n_estimators", 300)
         if isinstance(n_estimators_final, int):
             n_estimators_minimum = min(40, n_estimators_final)

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon/tabular/models/xgboost/callbacks.py

@@ -147,7 +147,7 @@ class EarlyStoppingCustom(EarlyStopping):
 
         model_size_memory_ratio = estimated_model_size_mb / available_mb
 
-        if (model_size_memory_ratio > 1.0) or (available_mb < 512):
+        if (model_size_memory_ratio > 0.75) or (available_mb < 512):
             logger.warning("Warning: Large XGB model size may cause OOM error if training continues")
             logger.warning(f"Available Memory: {available_mb} MB")
             logger.warning(f"Estimated XGB model size: {estimated_model_size_mb} MB")

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon/tabular/models/xgboost/xgboost_model.py

@@ -256,12 +256,25 @@ class XGBoostModel(AbstractModel):
         num_classes: int = 1,
         **kwargs,
     ) -> int:
+        if hyperparameters is None:
+            hyperparameters = {}
         num_classes = num_classes if num_classes else 1  # self.num_classes could be None after initialization if it's a regression problem
         data_mem_usage = get_approximate_df_mem_usage(X).sum()
         data_mem_usage_bytes = data_mem_usage * 7 + data_mem_usage / 4 * num_classes  # TODO: Extremely crude approximation, can be vastly improved
 
         max_bin = hyperparameters.get("max_bin", 256)
         max_depth = hyperparameters.get("max_depth", 6)
+        max_leaves = hyperparameters.get("max_leaves", 0)
+        if max_leaves is None:
+            max_leaves = 0
+
+        if max_depth > 12 or max_depth == 0:  # 0 = uncapped
+            max_depth = 12  # Try our best if the value is very large, only treat it as 12.
+
+        if max_leaves != 0:  # if capped max_leaves
+            # make the effective max_depth for calculations be the lesser of the two constraints
+            max_depth = min(max_depth, math.ceil(math.log2(max_leaves)))
+
         # Formula based on manual testing, aligns with LightGBM histogram sizes
         #  This approximation is less accurate than it is for LightGBM and CatBoost.
         #  Note that max_depth didn't appear to reduce memory usage below 6, and it was unclear if it increased memory usage above 6.
@@ -274,7 +287,12 @@ class XGBoostModel(AbstractModel):
         histogram_mem_usage_bytes = 20 * depth_modifier * len(X.columns) * max_bin
         histogram_mem_usage_bytes *= 1.2  # Add a 20% buffer
 
-        approx_mem_size_req = data_mem_usage_bytes + histogram_mem_usage_bytes
+        mem_size_per_estimator = num_classes * max_depth * 500  # very rough estimate
+        n_estimators = hyperparameters.get("n_estimators", 10000)
+        n_estimators_min = min(n_estimators, 1000)
+        mem_size_estimators = n_estimators_min * mem_size_per_estimator  # memory estimate after fitting up to 1000 estimators
+
+        approx_mem_size_req = data_mem_usage_bytes + histogram_mem_usage_bytes + mem_size_estimators
         return approx_mem_size_req
 
     def _validate_fit_memory_usage(self, mem_error_threshold: float = 1.0, mem_warning_threshold: float = 0.75, mem_size_threshold: int = 1e9, **kwargs):

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon/tabular/version.py

@@ -1,4 +1,4 @@
 """This is the autogluon version file."""
 
-__version__ = "1.2.1b20250425"
+__version__ = "1.2.1b20250426"
 __lite__ = False

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon.tabular.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.tabular
-Version: 1.2.1b20250425
+Version: 1.2.1b20250426
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community

{autogluon.tabular-1.2.1b20250425 → autogluon.tabular-1.2.1b20250426}/src/autogluon.tabular.egg-info/requires.txt

@@ -3,20 +3,20 @@ scipy<1.16,>=1.5.4
 pandas<2.3.0,>=2.0.0
 scikit-learn<1.7.0,>=1.4.0
 networkx<4,>=3.0
-autogluon.core==1.2.1b20250425
-autogluon.features==1.2.1b20250425
+autogluon.core==1.2.1b20250426
+autogluon.features==1.2.1b20250426
 
 [all]
-catboost<1.3,>=1.2
-autogluon.core[all]==1.2.1b20250425
-torch<2.7,>=2.2
-xgboost<3.1,>=2.0
 einops<0.9,>=0.7
-huggingface_hub[torch]
+torch<2.7,>=2.2
 lightgbm<4.7,>=4.0
+huggingface_hub[torch]
 numpy<2.0.0,>=1.25
 fastai<2.9,>=2.3.1
 spacy<3.8
+xgboost<3.1,>=2.0
+autogluon.core[all]==1.2.1b20250426
+catboost<1.3,>=1.2
 
 [catboost]
 numpy<2.0.0,>=1.25
@@ -34,7 +34,7 @@ imodels<2.1.0,>=1.3.10
 lightgbm<4.7,>=4.0
 
 [ray]
-autogluon.core[all]==1.2.1b20250425
+autogluon.core[all]==1.2.1b20250426
 
 [skex]
 scikit-learn-intelex<2025.5,>=2024.0