autogluon.tabular 1.5.1b20260105__py3-none-any.whl → 1.5.1b20260117__py3-none-any.whl

autogluon/tabular/models/fastainn/tabular_nn_fastai.py

@@ -95,14 +95,17 @@ class NNFastAiTabularModel(AbstractModel):
         'early.stopping.min_delta': 0.0001,
         'early.stopping.patience': 10,
     """
+
     ag_key = "FASTAI"
     ag_name = "NeuralNetFastAI"
     ag_priority = 50
     # Increase priority for multiclass since neural networks
     # scale better than trees as a function of n_classes.
-    ag_priority_by_problem_type = MappingProxyType({
-        MULTICLASS: 95,
-    })
+    ag_priority_by_problem_type = MappingProxyType(
+        {
+            MULTICLASS: 95,
+        }
+    )
     seed_name = "random_seed"
 
     model_internals_file_name = "model-internals.pkl"
@@ -136,8 +139,15 @@ class NNFastAiTabularModel(AbstractModel):
 
         if self.problem_type in [REGRESSION, QUANTILE] and self.y_scaler is not None:
             y_norm = pd.Series(self.y_scaler.fit_transform(y.values.reshape(-1, 1)).reshape(-1))
-            y_val_norm = pd.Series(self.y_scaler.transform(y_val.values.reshape(-1, 1)).reshape(-1)) if y_val is not None else None
-            logger.log(0, f"Training with scaled targets: {self.y_scaler} - !!! NN training metric will be different from the final results !!!")
+            y_val_norm = (
+                pd.Series(self.y_scaler.transform(y_val.values.reshape(-1, 1)).reshape(-1))
+                if y_val is not None
+                else None
+            )
+            logger.log(
+                0,
+                f"Training with scaled targets: {self.y_scaler} - !!! NN training metric will be different from the final results !!!",
+            )
         else:
             y_norm = y
             y_val_norm = y_val
@@ -170,14 +180,20 @@ class NNFastAiTabularModel(AbstractModel):
                 unique_vals = X[self.cont_columns].nunique()
                 self.cont_columns = [c for c in self.cont_columns if unique_vals[c] > 1]
             if self.cont_columns:
-                self._cont_normalization = (np.array(X[self.cont_columns].mean()), np.array(X[self.cont_columns].std()))
+                self._cont_normalization = (
+                    np.array(X[self.cont_columns].mean()),
+                    np.array(X[self.cont_columns].std()),
+                )
 
             num_cat_cols_og = len(self.cat_columns)
             if self.cat_columns:
                 try:
                     X_stats = X[self.cat_columns].describe(include="all").T.reset_index()
                     cat_cols_to_drop = list(
-                        X_stats[(X_stats["unique"] > self.params.get("max_unique_categorical_values", 10000)) | (X_stats["unique"].isna())]["index"].values
+                        X_stats[
+                            (X_stats["unique"] > self.params.get("max_unique_categorical_values", 10000))
+                            | (X_stats["unique"].isna())
+                        ]["index"].values
                     )
                 except:
                     cat_cols_to_drop = []
@@ -187,7 +203,9 @@ class NNFastAiTabularModel(AbstractModel):
             num_cat_cols_use = len(self.cat_columns)
             logger.log(15, f"Using {num_cat_cols_use}/{num_cat_cols_og} categorical features")
 
-            nullable_numeric_features = self._feature_metadata.get_features(valid_raw_types=[R_FLOAT, R_DATETIME], invalid_special_types=[S_TEXT_SPECIAL])
+            nullable_numeric_features = self._feature_metadata.get_features(
+                valid_raw_types=[R_FLOAT, R_DATETIME], invalid_special_types=[S_TEXT_SPECIAL]
+            )
             self.columns_fills = dict()
             self._columns_fills_names = nullable_numeric_features
             for c in self._columns_fills_names:  # No need to do this for int features, int can't have null
@@ -227,7 +245,9 @@ class NNFastAiTabularModel(AbstractModel):
             df = df.copy()
         return df
 
-    def _fit(self, X, y, X_val=None, y_val=None, time_limit=None, num_cpus=None, num_gpus=0, sample_weight=None, **kwargs):
+    def _fit(
+        self, X, y, X_val=None, y_val=None, time_limit=None, num_cpus=None, num_gpus=0, sample_weight=None, **kwargs
+    ):
         try_import_fastai()
         import torch
         from fastai import torch_core
@@ -240,7 +260,10 @@ class NNFastAiTabularModel(AbstractModel):
         torch.set_num_threads(num_cpus)
         start_time = time.time()
         if sample_weight is not None:  # TODO: support
-            logger.log(15, "sample_weight not yet supported for NNFastAiTabularModel, this model will ignore them in training.")
+            logger.log(
+                15,
+                "sample_weight not yet supported for NNFastAiTabularModel, this model will ignore them in training.",
+            )
 
         params = self._get_model_params()
         self._num_cpus_infer = params.pop("_num_cpus_infer", 1)
@@ -341,13 +364,19 @@ class NNFastAiTabularModel(AbstractModel):
 
         fname = "model"
         save_callback = AgSaveModelCallback(
-            monitor=objective_func_name_to_monitor, comp=objective_optim_mode, fname=fname, best_epoch_stop=best_epoch_stop, with_opt=True
+            monitor=objective_func_name_to_monitor,
+            comp=objective_optim_mode,
+            fname=fname,
+            best_epoch_stop=best_epoch_stop,
+            with_opt=True,
         )
 
         if time_limit is not None:
             time_elapsed = time.time() - start_time
             time_left = time_limit - time_elapsed
-            if time_left <= time_limit * 0.7:  # if 30% of time was spent preprocessing, likely not enough time to train model
+            if (
+                time_left <= time_limit * 0.7
+            ):  # if 30% of time was spent preprocessing, likely not enough time to train model
                 raise TimeLimitExceeded
         else:
             time_left = None
@@ -371,7 +400,12 @@ class NNFastAiTabularModel(AbstractModel):
                     self.model.path = Path(temp_dir)
 
                     len_val = len(X_val) if X_val is not None else 0
-                    epochs = self._get_epochs_number(samples_num=len(X) + len_val, epochs=params["epochs"], batch_size=batch_size, time_left=time_left)
+                    epochs = self._get_epochs_number(
+                        samples_num=len(X) + len_val,
+                        epochs=params["epochs"],
+                        batch_size=batch_size,
+                        time_left=time_left,
+                    )
                     if epochs == 0:
                         # Stop early if there is not enough time to train a full epoch
                         raise TimeLimitExceeded
@@ -474,7 +508,9 @@ class NNFastAiTabularModel(AbstractModel):
                 objective_func_name = "pinball_loss"
             else:
                 objective_func_name = "log_loss"
-            logger.warning(f"Metric {stopping_metric.name} is not supported by this model - using {objective_func_name} instead")
+            logger.warning(
+                f"Metric {stopping_metric.name} is not supported by this model - using {objective_func_name} instead"
+            )
 
         nn_metric = metrics_map.get(objective_func_name, None)
 
@@ -482,7 +518,11 @@ class NNFastAiTabularModel(AbstractModel):
 
     def __get_objective_func_to_monitor(self, objective_func_name):
         monitor_obj_func = {
-            **{k: m.name if hasattr(m, "name") else m.__name__ for k, m in self.__get_metrics_map().items() if m is not None},
+            **{
+                k: m.name if hasattr(m, "name") else m.__name__
+                for k, m in self.__get_metrics_map().items()
+                if m is not None
+            },
             "log_loss": "valid_loss",
         }
         objective_func_name_to_monitor = objective_func_name
@@ -534,13 +574,14 @@ class NNFastAiTabularModel(AbstractModel):
         self.model = __model
         # Export model
         if self._load_model:
-            save_pkl.save_with_fn(self._model_internals_path, self.model, pickle_fn=lambda m, buffer: export(m, buffer), verbose=verbose)
+            save_pkl.save_with_fn(
+                self._model_internals_path, self.model, pickle_fn=lambda m, buffer: export(m, buffer), verbose=verbose
+            )
         self._load_model = None
         return path
 
     @classmethod
     def load(cls, path: str, reset_paths=True, verbose=True):
-
         from fastai.learner import load_learner
 
         model = super().load(path, reset_paths=reset_paths, verbose=verbose)
@@ -627,7 +668,13 @@ class NNFastAiTabularModel(AbstractModel):
 
     def _estimate_memory_usage(self, X: pd.DataFrame, **kwargs) -> int:
         hyperparameters = self._get_model_params()
-        return self.estimate_memory_usage_static(X=X, problem_type=self.problem_type, num_classes=self.num_classes, hyperparameters=hyperparameters, **kwargs)
+        return self.estimate_memory_usage_static(
+            X=X,
+            problem_type=self.problem_type,
+            num_classes=self.num_classes,
+            hyperparameters=hyperparameters,
+            **kwargs,
+        )
 
     @classmethod
     def _estimate_memory_usage_static(

autogluon/tabular/models/fasttext/fasttext_model.py

@@ -79,7 +79,9 @@ class FastTextModel(AbstractModel):
                 params["verbose"] = 2
 
         if sample_weight is not None:
-            logger.log(15, "sample_weight not yet supported for FastTextModel, this model will ignore them in training.")
+            logger.log(
+                15, "sample_weight not yet supported for FastTextModel, this model will ignore them in training."
+            )
 
         X = self.preprocess(X)
 

autogluon/tabular/models/image_prediction/image_predictor.py

@@ -24,6 +24,7 @@ class ImagePredictorModel(MultiModalPredictorModel):
     Additionally has special null image handling to improve performance in the presence of null images (aka image path of '')
         Note: null handling has not been compared to the built-in null handling of MultimodalPredictor yet.
     """
+
     ag_key = "AG_IMAGE_NN"
     ag_name = "ImagePredictor"
 
@@ -61,14 +62,18 @@ class ImagePredictorModel(MultiModalPredictorModel):
         X, y, X_val, y_val = super().preprocess_fit(X=X, y=y, X_val=X_val, y_val=y_val, **kwargs)
         X_features = list(X.columns)
         if len(X_features) != 1:
-            raise AssertionError(f"ImagePredictorModel only supports one image feature, but {len(X_features)} were given: {X_features}")
+            raise AssertionError(
+                f"ImagePredictorModel only supports one image feature, but {len(X_features)} were given: {X_features}"
+            )
         self._image_col_name = X_features[0]
         null_indices = X[self._image_col_name] == ""
 
         # TODO: Consider some kind of weighting of the two options so there isn't a harsh cutoff at 50
         # FIXME: What if all rows in a class are null? Will probably crash.
         if null_indices.sum() > 50:
-            self._dummy_pred_proba = self._compute_dummy_pred_proba(y[null_indices])  # FIXME: Do this one for better results
+            self._dummy_pred_proba = self._compute_dummy_pred_proba(
+                y[null_indices]
+            )  # FIXME: Do this one for better results
         else:
             # Not enough null to get a confident estimate of null label average, instead use all data average
             self._dummy_pred_proba = self._compute_dummy_pred_proba(y)

autogluon/tabular/models/knn/knn_model.py

@@ -24,6 +24,7 @@ class KNNModel(AbstractModel):
     """
     KNearestNeighbors model (scikit-learn): https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.KNeighborsClassifier.html
     """
+
     ag_key = "KNN"
     ag_name = "KNeighbors"
     ag_priority = 100
@@ -106,11 +107,19 @@ class KNNModel(AbstractModel):
         if time_limit is None or num_rows_max <= 10000:
             self.model = self._get_model_type()(**params).fit(X, y)
         else:
-            self.model = self._fit_with_samples(X=X, y=y, model_params=params, time_limit=time_limit - (time.time() - time_start))
+            self.model = self._fit_with_samples(
+                X=X, y=y, model_params=params, time_limit=time_limit - (time.time() - time_start)
+            )
 
     def _estimate_memory_usage(self, X: pd.DataFrame, **kwargs) -> int:
         hyperparameters = self._get_model_params()
-        return self.estimate_memory_usage_static(X=X, problem_type=self.problem_type, num_classes=self.num_classes, hyperparameters=hyperparameters, **kwargs)
+        return self.estimate_memory_usage_static(
+            X=X,
+            problem_type=self.problem_type,
+            num_classes=self.num_classes,
+            hyperparameters=hyperparameters,
+            **kwargs,
+        )
 
     @classmethod
     def _estimate_memory_usage_static(
@@ -120,12 +129,23 @@ class KNNModel(AbstractModel):
         **kwargs,
     ) -> int:
         model_size_bytes = 4 * X.shape[0] * X.shape[1]  # Assuming float32 types
-        expected_final_model_size_bytes = int(model_size_bytes * 3.6)  # Roughly what can be expected of the final KNN model in memory size
+        expected_final_model_size_bytes = int(
+            model_size_bytes * 3.6
+        )  # Roughly what can be expected of the final KNN model in memory size
         return expected_final_model_size_bytes
 
-    def _validate_fit_memory_usage(self, mem_error_threshold: float = 0.2, mem_warning_threshold: float = 0.15, mem_size_threshold: int = 1e7, **kwargs):
+    def _validate_fit_memory_usage(
+        self,
+        mem_error_threshold: float = 0.2,
+        mem_warning_threshold: float = 0.15,
+        mem_size_threshold: int = 1e7,
+        **kwargs,
+    ):
         return super()._validate_fit_memory_usage(
-            mem_error_threshold=mem_error_threshold, mem_warning_threshold=mem_warning_threshold, mem_size_threshold=mem_size_threshold, **kwargs
+            mem_error_threshold=mem_error_threshold,
+            mem_warning_threshold=mem_warning_threshold,
+            mem_size_threshold=mem_size_threshold,
+            **kwargs,
         )
 
     # TODO: Won't work for RAPIDS without modification
@@ -167,7 +187,17 @@ class KNNModel(AbstractModel):
         return y_oof_pred_proba
 
     # TODO: Consider making this fully generic and available to all models
-    def _fit_with_samples(self, X, y, model_params, time_limit, start_samples=10000, max_samples=None, sample_growth_factor=2, sample_time_growth_factor=8):
+    def _fit_with_samples(
+        self,
+        X,
+        y,
+        model_params,
+        time_limit,
+        start_samples=10000,
+        max_samples=None,
+        sample_growth_factor=2,
+        sample_time_growth_factor=8,
+    ):
         """
         Fit model with samples of the data repeatedly, gradually increasing the amount of data until time_limit is reached or all data is used.
 
@@ -243,11 +273,14 @@ class KNNModel(AbstractModel):
             time_limit_left = time_limit - (time_fit_end_sample - time_start)
             time_fit_sample = time_limit_left_prior - time_limit_left
             time_required_for_next = time_fit_sample * sample_time_growth_factor
-            logger.log(15, f"\t{round(time_fit_sample, 2)}s \t= Train Time (Using {samples}/{num_rows_max} rows) ({round(time_limit_left, 2)}s remaining time)")
+            logger.log(
+                15,
+                f"\t{round(time_fit_sample, 2)}s \t= Train Time (Using {samples}/{num_rows_max} rows) ({round(time_limit_left, 2)}s remaining time)",
+            )
             if time_required_for_next > time_limit_left and i != len(num_rows_samples) - 1:
                 logger.log(
                     20,
-                    f"\tNot enough time to train KNN model on all training rows. Fit {samples}/{num_rows_max} rows. (Training KNN model on {num_rows_samples[i+1]} rows is expected to take {round(time_required_for_next, 2)}s)",
+                    f"\tNot enough time to train KNN model on all training rows. Fit {samples}/{num_rows_max} rows. (Training KNN model on {num_rows_samples[i + 1]} rows is expected to take {round(time_required_for_next, 2)}s)",
                 )
                 break
         if idx is not None:

autogluon/tabular/models/lgb/callbacks.py

@@ -74,12 +74,15 @@ def early_stopping_custom(
 
     def _init(env):
         if not ignore_dart_warning:
-            enabled[0] = not any((boost_alias in env.params and env.params[boost_alias] == "dart") for boost_alias in ("boosting", "boosting_type", "boost"))
+            enabled[0] = not any(
+                (boost_alias in env.params and env.params[boost_alias] == "dart")
+                for boost_alias in ("boosting", "boosting_type", "boost")
+            )
         if not enabled[0]:
             warnings.warn("Early stopping is not available in dart mode")
             return
         if not env.evaluation_result_list:
-            raise ValueError("For early stopping, " "at least one dataset and eval metric is required for evaluation")
+            raise ValueError("For early stopping, at least one dataset and eval metric is required for evaluation")
 
         if verbose:
             msg = "Training until validation scores don't improve for {} rounds."
@@ -179,7 +182,9 @@ def early_stopping_custom(
         if not enabled[0]:
             return
         if train_loss_name is not None:
-            train_loss_evals = [eval for eval in env.evaluation_result_list if eval[0] == "train_set" and eval[1] == train_loss_name]
+            train_loss_evals = [
+                eval for eval in env.evaluation_result_list if eval[0] == "train_set" and eval[1] == train_loss_name
+            ]
             train_loss_val = train_loss_evals[0][2]
         else:
             train_loss_val = 0.0
@@ -194,7 +199,9 @@ def early_stopping_custom(
                 best_score_list[i] = env.evaluation_result_list
                 best_trainloss[i] = train_loss_val
             if reporter is not None:  # Report current best scores for iteration, used in HPO
-                if i == indices_to_check[0]:  # TODO: documentation needs to note that we assume 0th index is the 'official' validation performance metric.
+                if (
+                    i == indices_to_check[0]
+                ):  # TODO: documentation needs to note that we assume 0th index is the 'official' validation performance metric.
                     if cmp_op[i] == gt:
                         validation_perf = score
                     else:
@@ -214,7 +221,10 @@ def early_stopping_custom(
                     logger.log(
                         15,
                         "Early stopping, best iteration is:\n[%d]\t%s"
-                        % (best_iter[i] + 1, "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]])),
+                        % (
+                            best_iter[i] + 1,
+                            "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]]),
+                        ),
                     )
                 raise EarlyStopException(best_iter[i], best_score_list[i])
             elif (max_diff is not None) and (abs(score - best_score[i]) > max_diff):
@@ -224,7 +234,10 @@ def early_stopping_custom(
                     logger.log(
                         15,
                         "Early stopping, best iteration is:\n[%d]\t%s"
-                        % (best_iter[i] + 1, "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]])),
+                        % (
+                            best_iter[i] + 1,
+                            "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]]),
+                        ),
                     )
                 raise EarlyStopException(best_iter[i], best_score_list[i])
             if env.iteration == env.end_iteration - 1:
@@ -232,7 +245,10 @@ def early_stopping_custom(
                     logger.log(
                         15,
                         "Did not meet early stopping criterion. Best iteration is:\n[%d]\t%s"
-                        % (best_iter[i] + 1, "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]])),
+                        % (
+                            best_iter[i] + 1,
+                            "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]]),
+                        ),
                     )
                 raise EarlyStopException(best_iter[i], best_score_list[i])
             if verbose:
@@ -243,7 +259,10 @@ def early_stopping_custom(
                 logger.log(
                     20,
                     "Found manual stop file, early stopping. Best iteration is:\n[%d]\t%s"
-                    % (best_iter[i] + 1, "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]])),
+                    % (
+                        best_iter[i] + 1,
+                        "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]]),
+                    ),
                 )
                 raise EarlyStopException(best_iter[i], best_score_list[i])
         if time_limit:
@@ -255,7 +274,11 @@ def early_stopping_custom(
                     20,
                     "\tRan out of time, early stopping on iteration "
                     + str(env.iteration + 1)
-                    + ". Best iteration is:\n\t[%d]\t%s" % (best_iter[i] + 1, "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]])),
+                    + ". Best iteration is:\n\t[%d]\t%s"
+                    % (
+                        best_iter[i] + 1,
+                        "\t".join([_format_eval_result(x, show_stdv=False) for x in best_score_list[i]]),
+                    ),
                 )
                 raise EarlyStopException(best_iter[i], best_score_list[i])
 

autogluon/tabular/models/lgb/hyperparameters/searchspaces.py

@@ -19,7 +19,9 @@ def get_searchspace_multiclass_baseline():
     params = {
         "learning_rate": space.Real(lower=5e-3, upper=0.2, default=0.05, log=True),
         "feature_fraction": space.Real(lower=0.75, upper=1.0, default=1.0),
-        "min_data_in_leaf": space.Int(lower=2, upper=60, default=20),  # TODO: Use size of dataset to set upper, if row count is small upper should be small
+        "min_data_in_leaf": space.Int(
+            lower=2, upper=60, default=20
+        ),  # TODO: Use size of dataset to set upper, if row count is small upper should be small
         "num_leaves": space.Int(
             lower=16, upper=96, default=31
         ),  # TODO: Use row count and feature count to set this, the higher feature count the higher num_leaves upper