autogluon.tabular 1.5.0b20251228__py3-none-any.whl → 1.5.1b20260116__py3-none-any.whl

autogluon/tabular/registry/__init__.py

@@ -1,2 +1,2 @@
-from ._model_registry import ModelRegistry
 from ._ag_model_registry import ag_model_registry
+from ._model_registry import ModelRegistry

autogluon/tabular/registry/_ag_model_registry.py

@@ -4,7 +4,6 @@ from autogluon.core.models import (
     SimpleWeightedEnsembleModel,
 )
 
-from . import ModelRegistry
 from ..models import (
     BoostedRulesModel,
     CatBoostModel,
@@ -18,25 +17,25 @@ from ..models import (
     KNNModel,
     LGBModel,
     LinearModel,
+    MitraModel,
     MultiModalPredictorModel,
     NNFastAiTabularModel,
     PrepLGBModel,
     RealMLPModel,
+    RealTabPFNv2Model,
+    RealTabPFNv25Model,
     RFModel,
     RuleFitModel,
     TabDPTModel,
     TabICLModel,
     TabMModel,
     TabPFNMixModel,
-    MitraModel,
-    RealTabPFNv2Model,
-    RealTabPFNv25Model,
     TabularNeuralNetTorchModel,
     TextPredictorModel,
     XGBoostModel,
     XTModel,
 )
-
+from ._model_registry import ModelRegistry
 
 # When adding a new model officially to AutoGluon, the model class should be added to the bottom of this list.
 REGISTERED_MODEL_CLS_LST = [

autogluon/tabular/registry/_model_registry.py

@@ -35,6 +35,7 @@ class ModelRegistry:
     predictor.fit(..., hyperparameters={"MY_MODEL": ...})
     ```
     """
+
     def __init__(self, model_cls_list: list[Type[AbstractModel]] | None = None):
         if model_cls_list is None:
             model_cls_list = []

autogluon/tabular/testing/fit_helper.py

@@ -2,14 +2,16 @@ from __future__ import annotations
 
 import copy
 import os
-import pandas as pd
 import shutil
-import sys
 import subprocess
+import sys
 import textwrap
 import uuid
 from typing import Any, Type
 
+import numpy as np
+import pandas as pd
+
 from autogluon.common.utils.path_converter import PathConverter
 from autogluon.core.constants import BINARY, MULTICLASS, REGRESSION
 from autogluon.core.metrics import METRICS
@@ -17,19 +19,18 @@ from autogluon.core.models import AbstractModel, BaggedEnsembleModel
 from autogluon.core.stacked_overfitting.utils import check_stacked_overfitting_from_leaderboard
 from autogluon.core.testing.global_context_snapshot import GlobalContextSnapshot
 from autogluon.core.utils import download, generate_train_test_split_combined, infer_problem_type, unzip
-
 from autogluon.tabular import TabularDataset, TabularPredictor
 from autogluon.tabular.testing.generate_datasets import (
-    generate_toy_binary_dataset,
     generate_toy_binary_10_dataset,
+    generate_toy_binary_dataset,
+    generate_toy_multiclass_10_dataset,
+    generate_toy_multiclass_30_dataset,
     generate_toy_multiclass_dataset,
-    generate_toy_regression_dataset,
+    generate_toy_quantile_10_dataset,
     generate_toy_quantile_dataset,
     generate_toy_quantile_single_level_dataset,
-    generate_toy_multiclass_10_dataset,
     generate_toy_regression_10_dataset,
-    generate_toy_quantile_10_dataset,
-    generate_toy_multiclass_30_dataset,
+    generate_toy_regression_dataset,
 )
 
 
@@ -154,6 +155,7 @@ class FitHelper:
     """
     Helper functions to test and verify predictors and models when fit through TabularPredictor's API.
     """
+
     @staticmethod
     def fit_and_validate_dataset(
         dataset_name: str,
@@ -181,11 +183,14 @@ class FitHelper:
         deepcopy_fit_args: bool = True,
         verify_model_seed: bool = False,
         verify_load_wo_cuda: bool = False,
+        verify_single_prediction_equivalent_to_multi: bool = True,
     ) -> TabularPredictor:
         if compiler_configs is None:
             compiler_configs = {}
         directory_prefix = "./datasets/"
-        train_data, test_data, dataset_info = DatasetLoaderHelper.load_dataset(name=dataset_name, directory_prefix=directory_prefix)
+        train_data, test_data, dataset_info = DatasetLoaderHelper.load_dataset(
+            name=dataset_name, directory_prefix=directory_prefix
+        )
         label = dataset_info["label"]
         problem_type = dataset_info["problem_type"]
         _init_args = dict(
@@ -234,7 +239,7 @@ class FitHelper:
             scikit_api=scikit_api,
             min_cls_count_train=min_cls_count_train,
         )
-        
+
         ctx_after = GlobalContextSnapshot.capture()
         ctx_before.assert_unchanged(ctx_after)
 
@@ -249,6 +254,26 @@ class FitHelper:
         if predictor.can_predict_proba:
             pred_proba = predictor.predict_proba(test_data)
             predictor.evaluate_predictions(y_true=test_data[label], y_pred=pred_proba)
+
+            pred_proba_repeat = predictor.predict_proba(test_data)
+            are_close = np.isclose(pred_proba, pred_proba_repeat).all()
+            if not are_close:
+                raise AssertionError(
+                    "Predictions differ when predicting on the same data multiple times\n"
+                    f"First Predict:\n{pred_proba}\n"
+                    f"Second Predict:\n{pred_proba_repeat}\n"
+                )
+
+            pred_proba_1 = predictor.predict_proba(test_data.head(1))  # Verify model can predict on a single sample
+            if verify_single_prediction_equivalent_to_multi:
+                pred_proba_1_from_multi = pred_proba.head(1)
+                are_close = np.isclose(pred_proba_1, pred_proba_1_from_multi).all()
+                if not are_close:
+                    raise AssertionError(
+                        "Predictions differ when predicting a single sample vs predicting multiple samples\n"
+                        f"Single Sample:\n{pred_proba_1}\n"
+                        f"Multi Sample:\n{pred_proba_1_from_multi}\n"
+                    )
         else:
             try:
                 predictor.predict_proba(test_data)
@@ -278,7 +303,9 @@ class FitHelper:
             model_info = model.get_info()
             can_refit_full = model._get_tags()["can_refit_full"]
             if can_refit_full:
-                assert not model_info["val_in_fit"], f"val data must not be present in refit model if `can_refit_full=True`. Maybe an exception occurred?"
+                assert not model_info["val_in_fit"], (
+                    f"val data must not be present in refit model if `can_refit_full=True`. Maybe an exception occurred?"
+                )
             else:
                 assert model_info["val_in_fit"], f"val data must be present in refit model if `can_refit_full=False`"
         if verify_model_seed:
@@ -293,7 +320,9 @@ class FitHelper:
         if extra_info:
             lb_kwargs["extra_info"] = True
         lb = predictor.leaderboard(test_data, extra_metrics=extra_metrics, **lb_kwargs)
-        stacked_overfitting_assert(lb, predictor, expected_stacked_overfitting_at_val, expected_stacked_overfitting_at_test)
+        stacked_overfitting_assert(
+            lb, predictor, expected_stacked_overfitting_at_val, expected_stacked_overfitting_at_test
+        )
 
         predictor_load = predictor.load(path=predictor.path)
         predictor_load.predict(test_data)
@@ -301,6 +330,7 @@ class FitHelper:
         # TODO: This is expensive, only do this sparingly.
         if verify_load_wo_cuda:
             import torch
+
             if torch.cuda.is_available():
                 # Checks if the model is able to predict w/o CUDA.
                 # This verifies that a model artifact works on a CPU machine.
@@ -322,7 +352,9 @@ class FitHelper:
 
         assert os.path.realpath(save_path) == os.path.realpath(predictor.path)
         if delete_directory:
-            shutil.rmtree(save_path, ignore_errors=True)  # Delete AutoGluon output directory to ensure runs' information has been removed.
+            shutil.rmtree(
+                save_path, ignore_errors=True
+            )  # Delete AutoGluon output directory to ensure runs' information has been removed.
         return predictor
 
     @staticmethod
@@ -379,6 +411,7 @@ class FitHelper:
         raise_on_model_failure: bool = True,
         problem_types: list[str] | None = None,
         verify_model_seed: bool = True,
+        verify_single_prediction_equivalent_to_multi: bool = True,
         **kwargs,
     ):
         """
@@ -396,6 +429,7 @@ class FitHelper:
             If specified, checks the given problem_types.
             If None, checks `model_cls.supported_problem_types()`
         verify_model_seed: bool = True
+        verify_single_prediction_equivalent_to_multi: bool = True
         **kwargs
 
         Returns
@@ -476,6 +510,7 @@ class FitHelper:
                     extra_metrics=_extra_metrics,
                     raise_on_model_failure=raise_on_model_failure,
                     verify_model_seed=verify_model_seed,
+                    verify_single_prediction_equivalent_to_multi=verify_single_prediction_equivalent_to_multi,
                     **kwargs,
                 )
 
@@ -508,6 +543,7 @@ class FitHelper:
                         extra_metrics=_extra_metrics,
                         raise_on_model_failure=raise_on_model_failure,
                         verify_model_seed=verify_model_seed,
+                        verify_single_prediction_equivalent_to_multi=verify_single_prediction_equivalent_to_multi,
                         **kwargs,
                     )
 
@@ -519,11 +555,15 @@ def stacked_overfitting_assert(
     expected_stacked_overfitting_at_test: bool | None,
 ):
     if expected_stacked_overfitting_at_val is not None:
-        assert predictor._stacked_overfitting_occurred == expected_stacked_overfitting_at_val, "Expected stacked overfitting at val mismatch!"
+        assert predictor._stacked_overfitting_occurred == expected_stacked_overfitting_at_val, (
+            "Expected stacked overfitting at val mismatch!"
+        )
 
     if expected_stacked_overfitting_at_test is not None:
         stacked_overfitting = check_stacked_overfitting_from_leaderboard(lb)
-        assert stacked_overfitting == expected_stacked_overfitting_at_test, "Expected stacked overfitting at test mismatch!"
+        assert stacked_overfitting == expected_stacked_overfitting_at_test, (
+            "Expected stacked overfitting at test mismatch!"
+        )
 
 
 def _verify_model_seed(model: AbstractModel):

autogluon/tabular/testing/generate_datasets.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import pandas as pd
 from sklearn.datasets import make_blobs
 
-from autogluon.core.constants import BINARY, MULTICLASS, REGRESSION, QUANTILE
+from autogluon.core.constants import BINARY, MULTICLASS, QUANTILE, REGRESSION
 
 
 def generate_toy_binary_dataset():

autogluon/tabular/testing/model_fit_helper.py

@@ -9,7 +9,6 @@ from autogluon.core.data.label_cleaner import LabelCleaner
 from autogluon.core.models import AbstractModel, BaggedEnsembleModel
 from autogluon.core.utils import generate_train_test_split, infer_problem_type
 from autogluon.features.generators import AbstractFeatureGenerator, AutoMLPipelineFeatureGenerator
-
 from autogluon.tabular.testing.fit_helper import FitHelper
 
 
@@ -18,6 +17,7 @@ class ModelFitHelper:
     """
     Helper functions to test and verify models when fit outside TabularPredictor's API (aka as stand-alone models)
     """
+
     @staticmethod
     def fit_and_validate_dataset(
         dataset_name: str,
@@ -27,7 +27,9 @@ class ModelFitHelper:
         check_predict_children: bool = False,
     ) -> AbstractModel:
         directory_prefix = "./datasets/"
-        train_data, test_data, dataset_info = FitHelper.load_dataset(name=dataset_name, directory_prefix=directory_prefix)
+        train_data, test_data, dataset_info = FitHelper.load_dataset(
+            name=dataset_name, directory_prefix=directory_prefix
+        )
         label = dataset_info["label"]
         model, label_cleaner, feature_generator = ModelFitHelper.fit_dataset(
             train_data=train_data, model=model, label=label, fit_args=fit_args, sample_size=sample_size
@@ -39,10 +41,14 @@ class ModelFitHelper:
         X_test = feature_generator.transform(X_test)
 
         y_pred = model.predict(X_test)
-        assert isinstance(y_pred, np.ndarray), f"Expected np.ndarray as model.predict(X_test) output. Got: {y_pred.__class__}"
+        assert isinstance(y_pred, np.ndarray), (
+            f"Expected np.ndarray as model.predict(X_test) output. Got: {y_pred.__class__}"
+        )
 
         y_pred_proba = model.predict_proba(X_test)
-        assert isinstance(y_pred_proba, np.ndarray), f"Expected np.ndarray as model.predict_proba(X_test) output. Got: {y_pred.__class__}"
+        assert isinstance(y_pred_proba, np.ndarray), (
+            f"Expected np.ndarray as model.predict_proba(X_test) output. Got: {y_pred.__class__}"
+        )
         model.get_info()
 
         if check_predict_children: