autogluon.tabular 1.3.2b20250722__tar.gz → 1.4.0__tar.gz

{autogluon.tabular-1.3.2b20250722 → autogluon.tabular-1.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.tabular
-Version: 1.3.2b20250722
+Version: 1.4.0
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community
@@ -10,7 +10,7 @@ Project-URL: Bug Reports, https://github.com/autogluon/autogluon/issues
 Project-URL: Source, https://github.com/autogluon/autogluon/
 Project-URL: Contribute!, https://github.com/autogluon/autogluon/blob/master/CONTRIBUTING.md
 Platform: UNKNOWN
-Classifier: Development Status :: 4 - Beta
+Classifier: Development Status :: 5 - Production/Stable
 Classifier: Intended Audience :: Education
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Science/Research
@@ -49,6 +49,7 @@ Provides-Extra: skex
 Provides-Extra: imodels
 Provides-Extra: skl2onnx
 Provides-Extra: all
+Provides-Extra: tabarena
 Provides-Extra: tests
 License-File: ../LICENSE
 License-File: ../NOTICE
@@ -95,7 +96,7 @@ Build accurate end-to-end ML models in just 3 lines of code!
 
 ```python
 from autogluon.tabular import TabularPredictor
-predictor = TabularPredictor(label="class").fit("train.csv")
+predictor = TabularPredictor(label="class").fit("train.csv", presets="best")
 predictions = predictor.predict("test.csv")
 ```
 

{autogluon.tabular-1.3.2b20250722 → autogluon.tabular-1.4.0}/setup.py

@@ -17,6 +17,7 @@ ag = importlib.util.module_from_spec(spec)
 spec.loader.exec_module(ag)
 ###########################
 
+import copy
 import sys
 
 version = ag.load_version_file()
@@ -46,7 +47,7 @@ extras_require = {
         "xgboost>=2.0,<3.1",  # <{N+1} upper cap, where N is the latest released minor version
     ],
     "realmlp": [
-        "pytabkit>=1.5,<1.6",
+        "pytabkit>=1.6,<1.7",
     ],
     "fastai": [
         "spacy<3.9",
@@ -69,8 +70,9 @@ extras_require = {
         "loguru",
         "einx",
         "omegaconf",
+        "torch",
         "transformers",
-        # "flash-attn>2.6.3,<2.8", # TODO: flash-attn installation requires --no-build-isolation and torch, python and cuda version compatibility.
+        "huggingface_hub[torch]",
     ],
     "tabicl": [
         "tabicl>=0.1.3,<0.2",  # 0.1.3 added a major bug fix to multithreading.
@@ -111,24 +113,38 @@ else:
 
 # TODO: v1.0: Rename `all` to `core`, make `all` contain everything.
 all_requires = []
-# TODO: Consider adding 'skex' to 'all'
 for extra_package in [
     "lightgbm",
     "catboost",
     "xgboost",
     "fastai",
     "tabm",
-    "tabpfnmix",
-    "realmlp",
+    "mitra",
     "ray",
 ]:
     all_requires += extras_require[extra_package]
 all_requires = list(set(all_requires))
 extras_require["all"] = all_requires
 
+tabarena_requires = copy.deepcopy(all_requires)
+for extra_package in [
+    "tabicl",
+    "tabpfn",
+    "realmlp",
+]:
+    tabarena_requires += extras_require[extra_package]
+tabarena_requires = list(set(tabarena_requires))
+extras_require["tabarena"] = tabarena_requires
 
 test_requires = []
-for test_package in ["tabpfn", "imodels", "skl2onnx", "tabicl", "mitra"]:
+for test_package in [
+    "tabicl",  # Currently has unnecessary extra dependencies such as xgboost and wandb
+    "tabpfn",
+    "realmlp",  # Will consider to put as part of `all_requires` once part of a portfolio
+    "tabpfnmix",  # Refer to `mitra`, which is an improved version of `tabpfnmix`
+    "imodels",
+    "skl2onnx",
+]:
     test_requires += extras_require[test_package]
 extras_require["tests"] = test_requires
 install_requires = ag.get_dependency_version_ranges(install_requires)

{autogluon.tabular-1.3.2b20250722 → autogluon.tabular-1.4.0}/src/autogluon/tabular/configs/config_helper.py

@@ -326,7 +326,7 @@ class ConfigBuilder:
         """
         If True, only the best model and its ancestor models are saved in the outputted `predictor`. All other models are deleted.
             If you only care about deploying the most accurate predictor with the smallest file-size and no longer need any of the other trained models or functionality beyond prediction on new data, then set: `keep_only_best=True`, `save_space=True`.
-            This is equivalent to calling `predictor.delete_models(models_to_keep='best', dry_run=False)` directly after `fit()`.
+            This is equivalent to calling `predictor.delete_models(models_to_keep='best')` directly after `fit()`.
         If used with `refit_full` and `set_best_to_refit_full`, the best model will be the refit_full model, and the original bagged best model will be deleted.
             `refit_full` will be automatically set to 'best' in this case to avoid training models which will be later deleted.
         """

autogluon.tabular-1.4.0/src/autogluon/tabular/configs/hyperparameter_configs.py

@@ -0,0 +1,144 @@
+import copy
+
+from .zeroshot.zeroshot_portfolio_2023 import hyperparameter_portfolio_zeroshot_2023
+from .zeroshot.zeroshot_portfolio_2025 import hyperparameter_portfolio_zeroshot_2025_small
+
+# Dictionary of preset hyperparameter configurations.
+hyperparameter_config_dict = dict(
+    # Default AutoGluon hyperparameters intended to maximize accuracy without significant regard to inference time or disk usage.
+    default={
+        "NN_TORCH": {},
+        "GBM": [
+            {"extra_trees": True, "ag_args": {"name_suffix": "XT"}},
+            {},
+            {
+                "learning_rate": 0.03,
+                "num_leaves": 128,
+                "feature_fraction": 0.9,
+                "min_data_in_leaf": 3,
+                "ag_args": {"name_suffix": "Large", "priority": 0, "hyperparameter_tune_kwargs": None},
+            },
+        ],
+        "CAT": {},
+        "XGB": {},
+        "FASTAI": {},
+        "RF": [
+            {"criterion": "gini", "ag_args": {"name_suffix": "Gini", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "entropy", "ag_args": {"name_suffix": "Entr", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "squared_error", "ag_args": {"name_suffix": "MSE", "problem_types": ["regression", "quantile"]}},
+        ],
+        "XT": [
+            {"criterion": "gini", "ag_args": {"name_suffix": "Gini", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "entropy", "ag_args": {"name_suffix": "Entr", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "squared_error", "ag_args": {"name_suffix": "MSE", "problem_types": ["regression", "quantile"]}},
+        ],
+    },
+    # Results in smaller models. Generally will make inference speed much faster and disk usage much lower, but with worse accuracy.
+    light={
+        "NN_TORCH": {},
+        "GBM": [
+            {"extra_trees": True, "ag_args": {"name_suffix": "XT"}},
+            {},
+            {
+                "learning_rate": 0.03,
+                "num_leaves": 128,
+                "feature_fraction": 0.9,
+                "min_data_in_leaf": 3,
+                "ag_args": {"name_suffix": "Large", "priority": 0, "hyperparameter_tune_kwargs": None},
+            },
+        ],
+        "CAT": {},
+        "XGB": {},
+        "FASTAI": {},
+        "RF": [
+            {"criterion": "gini", "max_depth": 15, "ag_args": {"name_suffix": "Gini", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "entropy", "max_depth": 15, "ag_args": {"name_suffix": "Entr", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "squared_error", "max_depth": 15, "ag_args": {"name_suffix": "MSE", "problem_types": ["regression", "quantile"]}},
+        ],
+        "XT": [
+            {"criterion": "gini", "max_depth": 15, "ag_args": {"name_suffix": "Gini", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "entropy", "max_depth": 15, "ag_args": {"name_suffix": "Entr", "problem_types": ["binary", "multiclass"]}},
+            {"criterion": "squared_error", "max_depth": 15, "ag_args": {"name_suffix": "MSE", "problem_types": ["regression", "quantile"]}},
+        ],
+    },
+    # Results in much smaller models. Behaves similarly to 'light', but in many cases with over 10x less disk usage and a further reduction in accuracy.
+    very_light={
+        "NN_TORCH": {},
+        "GBM": [
+            {"extra_trees": True, "ag_args": {"name_suffix": "XT"}},
+            {},
+            {
+                "learning_rate": 0.03,
+                "num_leaves": 128,
+                "feature_fraction": 0.9,
+                "min_data_in_leaf": 3,
+                "ag_args": {"name_suffix": "Large", "priority": 0, "hyperparameter_tune_kwargs": None},
+            },
+        ],
+        "CAT": {},
+        "XGB": {},
+        "FASTAI": {},
+    },
+    # Results in extremely quick to train models. Only use this when prototyping, as the model accuracy will be severely reduced.
+    toy={
+        "NN_TORCH": {"num_epochs": 5},
+        "GBM": {"num_boost_round": 10},
+        "CAT": {"iterations": 10},
+        "XGB": {"n_estimators": 10},
+    },
+    # Default AutoGluon hyperparameters intended to maximize accuracy in multimodal tabular + text datasets. Requires GPU.
+    multimodal={
+        "NN_TORCH": {},
+        "GBM": [
+            {},
+            {"extra_trees": True, "ag_args": {"name_suffix": "XT"}},
+            {
+                "learning_rate": 0.03,
+                "num_leaves": 128,
+                "feature_fraction": 0.9,
+                "min_data_in_leaf": 3,
+                "ag_args": {"name_suffix": "Large", "priority": 0, "hyperparameter_tune_kwargs": None},
+            },
+        ],
+        "CAT": {},
+        "XGB": {},
+        # 'FASTAI': {},  # FastAI gets killed if the dataset is large (400K rows).
+        "AG_AUTOMM": {},
+    },
+    # Hyperparameters intended to find an interpretable model which doesn't sacrifice predictive accuracy
+    interpretable={
+        "IM_RULEFIT": [{"max_rules": 7}, {"max_rules": 12}, {"max_rules": 18}],
+        "IM_FIGS": [{"max_rules": 6}, {"max_rules": 10}, {"max_rules": 15}],
+        # Note: Below are commented out because they are not meaningfully interpretable via the existing API
+        # 'IM_GREEDYTREE': [{'max_leaf_nodes': 7, 'max_leaf_nodes': 18}],
+        # 'IM_BOOSTEDRULES': [{'n_estimators': 5}, {'n_estimators': 10}],
+        # 'IM_HSTREE': [{'max_rules': 6}, {'max_rules': 12}, {'max_rules': 18}],
+    },
+    zeroshot=hyperparameter_portfolio_zeroshot_2023,
+    zeroshot_2023=hyperparameter_portfolio_zeroshot_2023,
+    zeroshot_2025_tabfm=hyperparameter_portfolio_zeroshot_2025_small,
+)
+
+tabpfnmix_default = {
+    "model_path_classifier": "autogluon/tabpfn-mix-1.0-classifier",
+    "model_path_regressor": "autogluon/tabpfn-mix-1.0-regressor",
+    "n_ensembles": 1,
+    "max_epochs": 30,
+    "ag.sample_rows_val": 5000,  # Beyond 5k val rows fine-tuning becomes very slow
+    "ag.max_rows": 50000,  # Beyond 50k rows, the time taken is longer than most users would like (hours), while the model is very weak at this size
+    "ag_args": {"name_suffix": "_v1"},
+}
+
+hyperparameter_config_dict["experimental_2024"] = {"TABPFNMIX": tabpfnmix_default}
+hyperparameter_config_dict["experimental_2024"].update(hyperparameter_config_dict["zeroshot_2023"])
+hyperparameter_config_dict["experimental"] = hyperparameter_config_dict["experimental_2024"]
+
+def get_hyperparameter_config_options():
+    return list(hyperparameter_config_dict.keys())
+
+
+def get_hyperparameter_config(config_name):
+    config_options = get_hyperparameter_config_options()
+    if config_name not in config_options:
+        raise ValueError(f"Valid hyperparameter config names are: {config_options}, but '{config_name}' was given instead.")
+    return copy.deepcopy(hyperparameter_config_dict[config_name])

{autogluon.tabular-1.3.2b20250722 → autogluon.tabular-1.4.0}/src/autogluon/tabular/configs/presets_configs.py

@@ -1,25 +1,5 @@
 # Dictionary of preset fit() parameter configurations.
 tabular_presets_dict = dict(
-    # [EXPERIMENTAL PRESET] The `experimental_quality` preset may be changed or removed without warning.
-    # This preset acts as a testing ground for cutting edge features and models which could later be added to the `best_quality` preset in future releases.
-    # Using this preset can lead to unexpected crashes, as it hasn't been as thoroughly tested as other presets.
-    # Absolute best predictive accuracy with **zero** consideration to inference time or disk usage.
-    # Recommended for applications that benefit from the best possible model accuracy and **do not** care about inference speed.
-    # Significantly stronger than `best_quality`, but can be over 10x slower in inference.
-    # Uses pre-trained tabular foundation models, which add a minimum of 1-2 GB to the predictor artifact's size.
-    # For best results, use as large of an instance as possible with as many CPU cores as possible (ideally 64+ cores)
-    # DOES NOT SUPPORT GPU.
-    # Aliases: experimental
-    experimental_quality={
-        "auto_stack": True,
-        "dynamic_stacking": "auto",
-        "num_bag_sets": 1,
-        "hyperparameters": "experimental",
-        "fit_strategy": "parallel",
-        "num_gpus": 0,
-        "time_limit": 3600,
-    },
-
     # Best predictive accuracy with little consideration to inference time or disk usage. Achieve even better results by specifying a large time_limit value.
     # Recommended for applications that benefit from the best possible model accuracy.
     # Aliases: best
@@ -94,7 +74,52 @@ tabular_presets_dict = dict(
     # ------------------------------------------
     # Experimental presets. Only use these presets if you are ok with unstable and potentially poor performing presets.
     #  Experimental presets can be removed or changed without warning.
-    # No Experimental Presets in v1.0
+
+    # [EXPERIMENTAL PRESET] The `extreme` preset may be changed or removed without warning.
+    # This preset acts as a testing ground for cutting edge features and models which could later be added to the `best_quality` preset in future releases.
+    # Using this preset can lead to unexpected crashes, as it hasn't been as thoroughly tested as other presets.
+    # Absolute best predictive accuracy with **zero** consideration to inference time or disk usage.
+    # Recommended for applications that benefit from the best possible model accuracy and **do not** care about inference speed.
+    # Significantly stronger than `best_quality`, but can be over 10x slower in inference.
+    # Uses pre-trained tabular foundation models, which add a minimum of 1-2 GB to the predictor artifact's size.
+    # For best results, use as large of an instance as possible with a GPU and as many CPU cores as possible (ideally 64+ cores)
+    # Aliases: extreme, experimental, experimental_quality
+    # GPU STRONGLY RECOMMENDED
+    extreme_quality={
+        "auto_stack": True,
+        "dynamic_stacking": "auto",
+        "num_bag_sets": 1,
+        "_experimental_dynamic_hyperparameters": True,
+        "hyperparameters": None,
+        "time_limit": 3600,
+    },
+
+    # Preset with a portfolio learned from TabArena v0.1: https://tabarena.ai/
+    # Uses tabular foundation models: TabPFNv2, TabICL, Mitra
+    # Uses deep learning model: TabM
+    # Uses tree models: LightGBM, CatBoost, XGBoost
+    # Extremely powerful on small datasets with <= 10000 training samples.
+    # Requires a GPU for best results.
+    tabarena={
+        "auto_stack": True,
+        "dynamic_stacking": "auto",
+        "num_bag_sets": 1,
+        "num_stack_levels": 0,
+        "hyperparameters": "zeroshot_2025_tabfm",
+        "time_limit": 3600,
+    },
+
+    # DOES NOT SUPPORT GPU.
+    experimental_quality_v120={
+        "auto_stack": True,
+        "dynamic_stacking": "auto",
+        "num_bag_sets": 1,
+        "hyperparameters": "experimental",
+        "fit_strategy": "parallel",
+        "num_gpus": 0,
+        "time_limit": 3600,
+    },
+
     # ------------------------------------------
     # ------------------------------------------
     # ------------------------------------------
@@ -103,7 +128,7 @@ tabular_presets_dict = dict(
 
 # Alias preset name alternatives
 tabular_presets_alias = dict(
-    experimental="experimental_quality",
+    extreme="extreme_quality",
     best="best_quality",
     high="high_quality",
     high_quality_fast_inference_only_refit="high_quality",
@@ -111,9 +136,12 @@ tabular_presets_alias = dict(
     good_quality_faster_inference_only_refit="good_quality",
     medium="medium_quality",
     medium_quality_faster_train="medium_quality",
-    eq="experimental_quality",
+    eq="extreme_quality",
     bq="best_quality",
     hq="high_quality",
     gq="good_quality",
     mq="medium_quality",
+    experimental="extreme_quality",
+    experimental_quality="extreme_quality",
+    experimental_quality_v140="extreme_quality",
 )

{autogluon.tabular-1.3.2b20250722 → autogluon.tabular-1.4.0}/src/autogluon/tabular/configs/zeroshot/zeroshot_portfolio_2023.py

@@ -793,5 +793,4 @@ hyperparameter_portfolio_zeroshot_2023 = {
         {"max_features": 1.0, "max_leaf_nodes": 40459, "min_samples_leaf": 1, "ag_args": {"name_suffix": "_r197", "priority": -78}},
         {"max_features": "sqrt", "max_leaf_nodes": 29702, "min_samples_leaf": 2, "ag_args": {"name_suffix": "_r126", "priority": -86}},
     ],
-    "KNN": [{"weights": "uniform", "ag_args": {"name_suffix": "Unif"}}, {"weights": "distance", "ag_args": {"name_suffix": "Dist"}}],
 }