FLAML 2.6.0__tar.gz → 2.7.0__tar.gz

{flaml-2.6.0 → flaml-2.7.0}/FLAML.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: FLAML
-Version: 2.6.0
+Version: 2.7.0
 Summary: A fast library for automated machine learning and tuning
 Author-email: Microsoft Corporation <hpo@microsoft.com>
 License: MIT
@@ -31,8 +31,8 @@ Requires-Dist: pandas<3; extra == "spark"
 Requires-Dist: joblibspark>=0.5.0; extra == "spark"
 Requires-Dist: joblib<=1.3.2; extra == "spark"
 Provides-Extra: test
-Requires-Dist: numpy<2.0.0,>=1.17; python_version < "3.13" and extra == "test"
-Requires-Dist: numpy>=1.17; python_version >= "3.13" and extra == "test"
+Requires-Dist: numpy<2.0.0,>=1.17; (python_version < "3.13" and (sys_platform != "win32" or platform_machine != "ARM64")) and extra == "test"
+Requires-Dist: numpy>=1.17; (python_version >= "3.13" or (sys_platform == "win32" and platform_machine == "ARM64")) and extra == "test"
 Requires-Dist: jupyter; extra == "test"
 Requires-Dist: lightgbm>=2.3.1; extra == "test"
 Requires-Dist: xgboost<2.0.0,>=0.90; python_version < "3.11" and extra == "test"
@@ -40,34 +40,37 @@ Requires-Dist: xgboost>=2.0.0; python_version >= "3.11" and extra == "test"
 Requires-Dist: scipy>=1.4.1; extra == "test"
 Requires-Dist: pandas>=1.1.4; extra == "test"
 Requires-Dist: scikit-learn>=1.2.0; extra == "test"
-Requires-Dist: thop; extra == "test"
+Requires-Dist: thop; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: pytest>=6.1.1; extra == "test"
 Requires-Dist: pytest-rerunfailures>=13.0; extra == "test"
+Requires-Dist: pytest-asyncio; (sys_platform == "win32" and platform_machine == "ARM64") and extra == "test"
 Requires-Dist: coverage>=5.3; extra == "test"
 Requires-Dist: pre-commit; extra == "test"
-Requires-Dist: torch; extra == "test"
-Requires-Dist: torchvision; extra == "test"
-Requires-Dist: catboost>=0.26; extra == "test"
+Requires-Dist: torch; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: torchvision; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: catboost>=0.26; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: rgf-python; extra == "test"
 Requires-Dist: optuna<=3.6.1,>=2.8.0; extra == "test"
-Requires-Dist: openml; extra == "test"
+Requires-Dist: openml; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: statsmodels>=0.12.2; extra == "test"
 Requires-Dist: psutil; extra == "test"
-Requires-Dist: transformers[torch]; extra == "test"
-Requires-Dist: datasets; extra == "test"
-Requires-Dist: evaluate; extra == "test"
+Requires-Dist: transformers[torch]; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: transformers; (sys_platform == "win32" and platform_machine == "ARM64") and extra == "test"
+Requires-Dist: datasets; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: evaluate; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: nltk!=3.8.2; extra == "test"
 Requires-Dist: rouge_score; extra == "test"
 Requires-Dist: hcrystalball; extra == "test"
 Requires-Dist: seqeval; extra == "test"
-Requires-Dist: pytorch-forecasting; extra == "test"
+Requires-Dist: pytorch-forecasting; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: cryptography==46.0.3; (sys_platform == "win32" and platform_machine == "ARM64") and extra == "test"
 Requires-Dist: mlflow-skinny<=2.22.1; extra == "test"
 Requires-Dist: joblibspark>=0.5.0; extra == "test"
 Requires-Dist: joblib<=1.3.2; extra == "test"
 Requires-Dist: nbconvert; extra == "test"
 Requires-Dist: nbformat; extra == "test"
 Requires-Dist: ipykernel; extra == "test"
-Requires-Dist: pytorch-lightning; extra == "test"
+Requires-Dist: pytorch-lightning; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: tensorboardX; extra == "test"
 Requires-Dist: requests; extra == "test"
 Requires-Dist: packaging; extra == "test"
@@ -142,6 +145,10 @@ Dynamic: license-file
     <br>
 </p>
 
+<p align="center">
+  <b>English</b> · <a href="https://github.com/microsoft/FLAML/blob/main/README.zh.md">简体中文</a>
+</p>
+
 :fire: FLAML supports AutoML and Hyperparameter Tuning in [Microsoft Fabric Data Science](https://learn.microsoft.com/en-us/fabric/data-science/automated-machine-learning-fabric). In addition, we've introduced Python 3.11+ support, along with a range of new estimators, and comprehensive integration with MLflow—thanks to contributions from the Microsoft Fabric product team.
 
 :fire: Heads-up: [AutoGen](https://microsoft.github.io/autogen/) has moved to a dedicated [GitHub repository](https://github.com/microsoft/autogen). FLAML no longer includes the `autogen` module—please use AutoGen directly.
@@ -202,7 +209,7 @@ automl.fit(X_train, y_train, task="classification", estimator_list=["lgbm"])
 from flaml import tune
 
 tune.run(
-    evaluation_function, config={…}, low_cost_partial_config={…}, time_budget_s=3600
+    evaluation_function, config={...}, low_cost_partial_config={...}, time_budget_s=3600
 )
 ```
 
@@ -237,7 +244,7 @@ This project welcomes contributions and suggestions. Most contributions require
 Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us
 the rights to use your contribution. For details, visit <https://cla.opensource.microsoft.com>.
 
-If you are new to GitHub [here](https://help.github.com/categories/collaborating-with-issues-and-pull-requests/) is a detailed help source on getting involved with development on GitHub.
+If you are new to GitHub [here](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests) is a detailed help source on getting involved with development on GitHub.
 
 When you submit a pull request, a CLA bot will automatically determine whether you need to provide
 a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions

{flaml-2.6.0 → flaml-2.7.0}/FLAML.egg-info/requires.txt

@@ -77,34 +77,24 @@ lightgbm>=2.3.1
 scipy>=1.4.1
 pandas>=1.1.4
 scikit-learn>=1.2.0
-thop
 pytest>=6.1.1
 pytest-rerunfailures>=13.0
 coverage>=5.3
 pre-commit
-torch
-torchvision
-catboost>=0.26
 rgf-python
 optuna<=3.6.1,>=2.8.0
-openml
 statsmodels>=0.12.2
 psutil
-transformers[torch]
-datasets
-evaluate
 nltk!=3.8.2
 rouge_score
 hcrystalball
 seqeval
-pytorch-forecasting
 mlflow-skinny<=2.22.1
 joblibspark>=0.5.0
 joblib<=1.3.2
 nbconvert
 nbformat
 ipykernel
-pytorch-lightning
 tensorboardX
 requests
 packaging
@@ -113,15 +103,32 @@ dill
 [test:python_version < "3.11"]
 xgboost<2.0.0,>=0.90
 
-[test:python_version < "3.13"]
+[test:python_version < "3.13" and (sys_platform != "win32" or platform_machine != "ARM64")]
 numpy<2.0.0,>=1.17
 
 [test:python_version >= "3.11"]
 xgboost>=2.0.0
 
-[test:python_version >= "3.13"]
+[test:python_version >= "3.13" or (sys_platform == "win32" and platform_machine == "ARM64")]
 numpy>=1.17
 
+[test:sys_platform != "win32" or platform_machine != "ARM64"]
+thop
+torch
+torchvision
+catboost>=0.26
+openml
+transformers[torch]
+datasets
+evaluate
+pytorch-forecasting
+pytorch-lightning
+
+[test:sys_platform == "win32" and platform_machine == "ARM64"]
+pytest-asyncio
+transformers
+cryptography==46.0.3
+
 [ts_forecast]
 holidays
 prophet>=1.1.5

{flaml-2.6.0 → flaml-2.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: FLAML
-Version: 2.6.0
+Version: 2.7.0
 Summary: A fast library for automated machine learning and tuning
 Author-email: Microsoft Corporation <hpo@microsoft.com>
 License: MIT
@@ -31,8 +31,8 @@ Requires-Dist: pandas<3; extra == "spark"
 Requires-Dist: joblibspark>=0.5.0; extra == "spark"
 Requires-Dist: joblib<=1.3.2; extra == "spark"
 Provides-Extra: test
-Requires-Dist: numpy<2.0.0,>=1.17; python_version < "3.13" and extra == "test"
-Requires-Dist: numpy>=1.17; python_version >= "3.13" and extra == "test"
+Requires-Dist: numpy<2.0.0,>=1.17; (python_version < "3.13" and (sys_platform != "win32" or platform_machine != "ARM64")) and extra == "test"
+Requires-Dist: numpy>=1.17; (python_version >= "3.13" or (sys_platform == "win32" and platform_machine == "ARM64")) and extra == "test"
 Requires-Dist: jupyter; extra == "test"
 Requires-Dist: lightgbm>=2.3.1; extra == "test"
 Requires-Dist: xgboost<2.0.0,>=0.90; python_version < "3.11" and extra == "test"
@@ -40,34 +40,37 @@ Requires-Dist: xgboost>=2.0.0; python_version >= "3.11" and extra == "test"
 Requires-Dist: scipy>=1.4.1; extra == "test"
 Requires-Dist: pandas>=1.1.4; extra == "test"
 Requires-Dist: scikit-learn>=1.2.0; extra == "test"
-Requires-Dist: thop; extra == "test"
+Requires-Dist: thop; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: pytest>=6.1.1; extra == "test"
 Requires-Dist: pytest-rerunfailures>=13.0; extra == "test"
+Requires-Dist: pytest-asyncio; (sys_platform == "win32" and platform_machine == "ARM64") and extra == "test"
 Requires-Dist: coverage>=5.3; extra == "test"
 Requires-Dist: pre-commit; extra == "test"
-Requires-Dist: torch; extra == "test"
-Requires-Dist: torchvision; extra == "test"
-Requires-Dist: catboost>=0.26; extra == "test"
+Requires-Dist: torch; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: torchvision; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: catboost>=0.26; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: rgf-python; extra == "test"
 Requires-Dist: optuna<=3.6.1,>=2.8.0; extra == "test"
-Requires-Dist: openml; extra == "test"
+Requires-Dist: openml; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: statsmodels>=0.12.2; extra == "test"
 Requires-Dist: psutil; extra == "test"
-Requires-Dist: transformers[torch]; extra == "test"
-Requires-Dist: datasets; extra == "test"
-Requires-Dist: evaluate; extra == "test"
+Requires-Dist: transformers[torch]; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: transformers; (sys_platform == "win32" and platform_machine == "ARM64") and extra == "test"
+Requires-Dist: datasets; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: evaluate; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: nltk!=3.8.2; extra == "test"
 Requires-Dist: rouge_score; extra == "test"
 Requires-Dist: hcrystalball; extra == "test"
 Requires-Dist: seqeval; extra == "test"
-Requires-Dist: pytorch-forecasting; extra == "test"
+Requires-Dist: pytorch-forecasting; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
+Requires-Dist: cryptography==46.0.3; (sys_platform == "win32" and platform_machine == "ARM64") and extra == "test"
 Requires-Dist: mlflow-skinny<=2.22.1; extra == "test"
 Requires-Dist: joblibspark>=0.5.0; extra == "test"
 Requires-Dist: joblib<=1.3.2; extra == "test"
 Requires-Dist: nbconvert; extra == "test"
 Requires-Dist: nbformat; extra == "test"
 Requires-Dist: ipykernel; extra == "test"
-Requires-Dist: pytorch-lightning; extra == "test"
+Requires-Dist: pytorch-lightning; (sys_platform != "win32" or platform_machine != "ARM64") and extra == "test"
 Requires-Dist: tensorboardX; extra == "test"
 Requires-Dist: requests; extra == "test"
 Requires-Dist: packaging; extra == "test"
@@ -142,6 +145,10 @@ Dynamic: license-file
     <br>
 </p>
 
+<p align="center">
+  <b>English</b> · <a href="https://github.com/microsoft/FLAML/blob/main/README.zh.md">简体中文</a>
+</p>
+
 :fire: FLAML supports AutoML and Hyperparameter Tuning in [Microsoft Fabric Data Science](https://learn.microsoft.com/en-us/fabric/data-science/automated-machine-learning-fabric). In addition, we've introduced Python 3.11+ support, along with a range of new estimators, and comprehensive integration with MLflow—thanks to contributions from the Microsoft Fabric product team.
 
 :fire: Heads-up: [AutoGen](https://microsoft.github.io/autogen/) has moved to a dedicated [GitHub repository](https://github.com/microsoft/autogen). FLAML no longer includes the `autogen` module—please use AutoGen directly.
@@ -202,7 +209,7 @@ automl.fit(X_train, y_train, task="classification", estimator_list=["lgbm"])
 from flaml import tune
 
 tune.run(
-    evaluation_function, config={…}, low_cost_partial_config={…}, time_budget_s=3600
+    evaluation_function, config={...}, low_cost_partial_config={...}, time_budget_s=3600
 )
 ```
 
@@ -237,7 +244,7 @@ This project welcomes contributions and suggestions. Most contributions require
 Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us
 the rights to use your contribution. For details, visit <https://cla.opensource.microsoft.com>.
 
-If you are new to GitHub [here](https://help.github.com/categories/collaborating-with-issues-and-pull-requests/) is a detailed help source on getting involved with development on GitHub.
+If you are new to GitHub [here](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests) is a detailed help source on getting involved with development on GitHub.
 
 When you submit a pull request, a CLA bot will automatically determine whether you need to provide
 a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions

{flaml-2.6.0 → flaml-2.7.0}/README.md

@@ -14,6 +14,10 @@
     <br>
 </p>
 
+<p align="center">
+  <b>English</b> · <a href="https://github.com/microsoft/FLAML/blob/main/README.zh.md">简体中文</a>
+</p>
+
 :fire: FLAML supports AutoML and Hyperparameter Tuning in [Microsoft Fabric Data Science](https://learn.microsoft.com/en-us/fabric/data-science/automated-machine-learning-fabric). In addition, we've introduced Python 3.11+ support, along with a range of new estimators, and comprehensive integration with MLflow—thanks to contributions from the Microsoft Fabric product team.
 
 :fire: Heads-up: [AutoGen](https://microsoft.github.io/autogen/) has moved to a dedicated [GitHub repository](https://github.com/microsoft/autogen). FLAML no longer includes the `autogen` module—please use AutoGen directly.
@@ -74,7 +78,7 @@ automl.fit(X_train, y_train, task="classification", estimator_list=["lgbm"])
 from flaml import tune
 
 tune.run(
-    evaluation_function, config={…}, low_cost_partial_config={…}, time_budget_s=3600
+    evaluation_function, config={...}, low_cost_partial_config={...}, time_budget_s=3600
 )
 ```
 
@@ -109,7 +113,7 @@ This project welcomes contributions and suggestions. Most contributions require
 Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us
 the rights to use your contribution. For details, visit <https://cla.opensource.microsoft.com>.
 
-If you are new to GitHub [here](https://help.github.com/categories/collaborating-with-issues-and-pull-requests/) is a detailed help source on getting involved with development on GitHub.
+If you are new to GitHub [here](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests) is a detailed help source on getting involved with development on GitHub.
 
 When you submit a pull request, a CLA bot will automatically determine whether you need to provide
 a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions

{flaml-2.6.0 → flaml-2.7.0}/flaml/automl/automl.py

@@ -19,7 +19,12 @@ import numpy as np
 
 from flaml import tune
 from flaml.automl.logger import logger, logger_formatter
-from flaml.automl.ml import huggingface_metric_to_mode, sklearn_metric_name_set, spark_metric_name_dict, train_estimator
+from flaml.automl.ml import (
+    huggingface_metric_to_mode,
+    sklearn_metric_name_set,
+    spark_metric_name_dict,
+    train_estimator,
+)
 from flaml.automl.spark import DataFrame, Series, psDataFrame, psSeries
 from flaml.automl.state import AutoMLState, SearchState
 from flaml.automl.task.factory import task_factory
@@ -285,6 +290,9 @@ class AutoML(BaseEstimator):
                 records to the input log file if it exists.
             auto_augment: boolean, default=True | Whether to automatically
                 augment rare classes.
+            resampler: object, default=None | An imbalanced-learn-compatible resampler
+                to apply to training partitions. It must be cloneable via
+                `sklearn.base.clone` and expose `fit_resample(X, y) -> (X, y)`.
             min_sample_size: int, default=MIN_SAMPLE_TRAIN | the minimal sample
                 size when sample=True.
             use_ray: boolean or dict.
@@ -419,6 +427,7 @@ class AutoML(BaseEstimator):
         settings["custom_hp"] = settings.get("custom_hp", {})
         settings["skip_transform"] = settings.get("skip_transform", False)
         settings["mlflow_logging"] = settings.get("mlflow_logging", True)
+        settings["resampler"] = settings.get("resampler")
 
         self._estimator_type = "classifier" if settings["task"] in CLASSIFICATION else "regressor"
         self.best_run_id = None
@@ -497,6 +506,51 @@ class AutoML(BaseEstimator):
                 f"(e.g., metric=custom_metric(...))."
             )
 
+    @staticmethod
+    def _validate_resampler(resampler, groups, ensemble, fit_kwargs, fit_kwargs_by_estimator):
+        if resampler is None:
+            return fit_kwargs, fit_kwargs_by_estimator
+        if ensemble:
+            raise ValueError(
+                "Cannot combine 'resampler' with 'ensemble' because stacking performs "
+                "internal cross-validation where pre-resampling can leak synthetic samples "
+                "across folds. Disable ensemble or omit resampler."
+            )
+        metadata_sources = [fit_kwargs] + list((fit_kwargs_by_estimator or {}).values())
+        if any(kwargs.get("sample_weight") is not None for kwargs in metadata_sources):
+            raise ValueError(
+                "Cannot combine 'resampler' with 'sample_weight' (including via "
+                "fit_kwargs_by_estimator) — resampling breaks the 1-to-1 row alignment "
+                "with sample weights. Use either resampling or sample weighting, not both."
+            )
+        if groups is not None or any(kwargs.get("groups") is not None for kwargs in metadata_sources):
+            raise ValueError(
+                "Cannot combine 'resampler' with 'groups' because resampling breaks the "
+                "1-to-1 row alignment with group labels."
+            )
+        if not callable(getattr(resampler, "fit_resample", None)):
+            raise TypeError(
+                "'resampler' must expose a fit_resample(X, y) -> (X, y) method "
+                "(e.g., an imbalanced-learn BaseSampler such as SMOTE)."
+            )
+        try:
+            from sklearn.base import clone
+
+            clone(resampler)
+        except Exception as e:
+            raise TypeError(
+                "'resampler' must be cloneable via sklearn.base.clone (implement "
+                "get_params/set_params, e.g. by subclassing sklearn.base.BaseEstimator); "
+                f"cloning failed with: {e}"
+            ) from e
+        return (
+            {key: value for key, value in fit_kwargs.items() if key != "sample_weight"},
+            {
+                name: {key: value for key, value in kwargs.items() if key != "sample_weight"}
+                for name, kwargs in (fit_kwargs_by_estimator or {}).items()
+            },
+        )
+
     def get_params(self, deep: bool = False) -> dict:
         return self._settings.copy()
 
@@ -890,6 +944,14 @@ class AutoML(BaseEstimator):
 
             # Apply task-level preprocessing to new data
             X_test_preprocessed = automl.preprocess(X_test)
+
+            # Required when calling a single ensemble component directly, since
+            # `automl.model.estimators_[i]` was fitted on already-preprocessed
+            # data and cannot consume raw input (see issue #1136):
+            automl_ensemble = AutoML()
+            automl_ensemble.fit(X_train, y_train, task="classification", ensemble=True)
+            X_test_preprocessed = automl_ensemble.preprocess(X_test)
+            component_pred = automl_ensemble.model.estimators_[0].predict(X_test_preprocessed)
             ```
         """
         if not hasattr(self, "_state") or self._state is None:
@@ -966,6 +1028,8 @@ class AutoML(BaseEstimator):
         skip_transform=None,
         preserve_checkpoint=True,
         fit_kwargs_by_estimator=None,
+        *,
+        resampler=None,
         **fit_kwargs,
     ):
         """Retrain from log file.
@@ -1025,6 +1089,9 @@ class AutoML(BaseEstimator):
                 when `record_id >= 0`, `time_budget` will be ignored.
             auto_augment: boolean, default=True | Whether to automatically
                 augment rare classes.
+            resampler: object, default=None | An imbalanced-learn-compatible resampler
+                to apply before retraining. It must be cloneable via `sklearn.base.clone`
+                and expose `fit_resample(X, y) -> (X, y)`.
             custom_hp: dict, default=None | The custom search space specified by user
                 Each key is the estimator name, each value is a dict of the custom search space for that estimator. Notice the
                 domain of the custom search space can either be a value or a sample.Domain object.
@@ -1095,6 +1162,9 @@ class AutoML(BaseEstimator):
         n_splits = n_splits or self._settings.get("n_splits")
         split_type = split_type or self._settings.get("split_type")
         auto_augment = self._settings.get("auto_augment") if auto_augment is None else auto_augment
+        resampler = self._settings.get("resampler") if resampler is None else resampler
+        if resampler is not None:
+            auto_augment = False
         self._state.task = task
         self._estimator_type = "classifier" if task.is_classification() else "regressor"
 
@@ -1102,6 +1172,11 @@ class AutoML(BaseEstimator):
         self._state.custom_hp = custom_hp or self._settings.get("custom_hp")
         self._skip_transform = self._settings.get("skip_transform") if skip_transform is None else skip_transform
         self._state.fit_kwargs_by_estimator = fit_kwargs_by_estimator or self._settings.get("fit_kwargs_by_estimator")
+        fit_kwargs, self._state.fit_kwargs_by_estimator = self._validate_resampler(
+            resampler, groups, False, fit_kwargs, self._state.fit_kwargs_by_estimator
+        )
+        self._state.fit_kwargs = fit_kwargs
+        task._resampler = resampler
         self.preserve_checkpoint = (
             self._settings.get("preserve_checkpoint") if preserve_checkpoint is None else preserve_checkpoint
         )
@@ -1837,6 +1912,8 @@ class AutoML(BaseEstimator):
         mlflow_logging=None,
         fit_kwargs_by_estimator=None,
         mlflow_exp_name=None,
+        *,
+        resampler=None,
         **fit_kwargs,
     ):
         """Find a model for a given task.
@@ -2155,6 +2232,17 @@ class AutoML(BaseEstimator):
         }
         ```
 
+            resampler: object, default=None | An imbalanced-learn-compatible resampler
+                (such as `imblearn.over_sampling.SMOTE`) that is cloneable via
+                `sklearn.base.clone` and exposes `fit_resample(X, y) -> (X, y)`. When set,
+                the resampler is cloned and applied to each cross-validation fold's training
+                partition, the holdout training partition, and final or retrain data.
+                Validation partitions are left at the raw class distribution, and automatic
+                rare-class augmentation is disabled. Automatically generated classification
+                holdout partitions are disjoint; singleton classes raise `ValueError`.
+                Not compatible with non-null `sample_weight`,
+                `groups`, or `ensemble`; passing these combinations raises `ValueError`.
+                Off by default. See issue #1200 for the design discussion and benchmarks.
             **fit_kwargs: Other key word arguments to pass to fit() function of
                 the searched learners, such as sample_weight. Below are a few examples of
                 estimator-specific parameters:
@@ -2203,6 +2291,9 @@ class AutoML(BaseEstimator):
         split_ratio = split_ratio or self._settings.get("split_ratio")
         n_splits = n_splits or self._settings.get("n_splits")
         auto_augment = self._settings.get("auto_augment") if auto_augment is None else auto_augment
+        resampler = self._settings.get("resampler") if resampler is None else resampler
+        if resampler is not None:
+            auto_augment = False
         allow_label_overlap = (
             self._settings.get("allow_label_overlap") if allow_label_overlap is None else allow_label_overlap
         )
@@ -2328,6 +2419,11 @@ class AutoML(BaseEstimator):
             self._state.resources_per_trial = {"cpu": n_jobs} if n_jobs > 0 else {"cpu": 1}
         self._state.free_mem_ratio = self._settings.get("free_mem_ratio") if free_mem_ratio is None else free_mem_ratio
         self._state.task = task
+        fit_kwargs_by_estimator = fit_kwargs_by_estimator or self._settings.get("fit_kwargs_by_estimator")
+        fit_kwargs, fit_kwargs_by_estimator = self._validate_resampler(
+            resampler, groups, ensemble, fit_kwargs, fit_kwargs_by_estimator
+        )
+        task._resampler = resampler
         self._state.log_training_metric = log_training_metric
 
         self._state.fit_kwargs = fit_kwargs
@@ -2340,7 +2436,6 @@ class AutoML(BaseEstimator):
             if mlflow_logging is None
             else mlflow_logging
         )
-        fit_kwargs_by_estimator = fit_kwargs_by_estimator or self._settings.get("fit_kwargs_by_estimator")
         self._state.fit_kwargs_by_estimator = fit_kwargs_by_estimator.copy()  # shallow copy of fit_kwargs_by_estimator
         self._state.weight_val = sample_weight_val
         self._mlflow_exp_name = mlflow_exp_name

{flaml-2.6.0 → flaml-2.7.0}/flaml/automl/data.py

@@ -7,6 +7,7 @@ import os
 import random
 import re
 import uuid
+import warnings
 from datetime import datetime, timedelta
 from decimal import ROUND_HALF_UP, Decimal
 from typing import TYPE_CHECKING, Union
@@ -350,6 +351,17 @@ class DataTransformer:
                 X.insert(0, TS_TIMESTAMP_COL, ds_col)
             if cat_columns:
                 X[cat_columns] = X[cat_columns].astype("category")
+                # Pin the per-column category list seen at fit time so
+                # `transform()` produces the same integer codes for the same
+                # values regardless of what is passed at predict time (see
+                # issue #1101). "__NAN__" is reserved as the sentinel slot
+                # used for values unseen at fit time.
+                self._cat_categories = {}
+                for col in cat_columns:
+                    cats = list(X[col].cat.categories)
+                    if "__NAN__" not in cats:
+                        cats.append("__NAN__")
+                    self._cat_categories[col] = cats
             if num_columns:
                 X_num = X[num_columns]
                 try:
@@ -450,6 +462,30 @@ class DataTransformer:
                         X[column] = X[column].cat.add_categories("__NAN__").fillna("__NAN__")
             if cat_columns:
                 X[cat_columns] = X[cat_columns].astype("category")
+                # Pin codes to the categories seen at fit time so they do not
+                # drift when the predict-time column has a different value
+                # distribution than the fit-time column (see issue #1101).
+                # Older pickles without `_cat_categories` fall back to
+                # whatever `astype("category")` inferred above.
+                saved_cats_map = getattr(self, "_cat_categories", None)
+                if saved_cats_map:
+                    for column in cat_columns:
+                        saved_cats = saved_cats_map.get(column)
+                        if saved_cats is None:
+                            continue
+                        current = X[column].astype(object)
+                        unseen_mask = ~current.isin(saved_cats) & current.notna()
+                        if unseen_mask.any():
+                            samples = sorted({str(v) for v in current[unseen_mask].unique()})[:5]
+                            warnings.warn(
+                                f"Column '{column}' contains values unseen at fit time "
+                                f"(e.g. {samples}); these rows will be encoded as '__NAN__' "
+                                "and predictions may be unreliable.",
+                                UserWarning,
+                                stacklevel=2,
+                            )
+                            current = current.where(~unseen_mask, "__NAN__")
+                        X[column] = pd.Categorical(current, categories=saved_cats)
             if num_columns:
                 X_num = X[num_columns].fillna(np.nan)
                 if self._drop:

{flaml-2.6.0 → flaml-2.7.0}/flaml/automl/ml.py

@@ -388,7 +388,7 @@ def compute_estimator(
             budget=budget,
             log_training_metric=log_training_metric,
             fit_kwargs=fit_kwargs,
-            free_mem_ratio=0,
+            free_mem_ratio=free_mem_ratio,
         )
     else:
         val_loss, metric_for_logging, train_time, pred_time = task.evaluate_model_CV(
@@ -403,7 +403,7 @@ def compute_estimator(
             cv_score_agg_func,
             log_training_metric=log_training_metric,
             fit_kwargs=fit_kwargs,
-            free_mem_ratio=0,
+            free_mem_ratio=free_mem_ratio,
         )
 
     if isinstance(estimator, TransformersEstimator):
@@ -448,6 +448,7 @@ def train_estimator(
         fit_kwargs["metric"] = eval_metric
 
     if X_train is not None:
+        X_train, y_train = _resample_training_data(X_train, y_train, task)
         train_time = estimator.fit(X_train, y_train, budget=budget, free_mem_ratio=free_mem_ratio, **fit_kwargs)
     else:
         estimator = estimator.estimator_class(**estimator.params)
@@ -455,6 +456,16 @@ def train_estimator(
     return estimator, train_time
 
 
+def _resample_training_data(X_train, y_train, task):
+    resampler = getattr(task, "_resampler", None)
+    if resampler is None:
+        return X_train, y_train
+
+    from sklearn.base import clone
+
+    return clone(resampler).fit_resample(X_train, y_train)
+
+
 def norm_confusion_matrix(y_true: Union[np.array, Series], y_pred: Union[np.array, Series]):
     """normalized confusion matrix.
 
@@ -525,6 +536,7 @@ def get_val_loss(
     #     fit_kwargs['groups_val'] = groups_val
     #     fit_kwargs['X_val'] = X_val
     #     fit_kwargs['y_val'] = y_val
+    X_train, y_train = _resample_training_data(X_train, y_train, task)
     estimator.fit(X_train, y_train, budget=budget, free_mem_ratio=free_mem_ratio, **fit_kwargs)
     val_loss, metric_for_logging, pred_time, _ = _eval_estimator(
         config,