perpetual 0.7.12__cp313-none-win_amd64.whl → 0.8.1__cp313-none-win_amd64.whl

perpetual/booster.py

@@ -26,7 +26,7 @@ class PerpetualBooster:
     # this is useful for parameters that should be
     # attempted to be loaded in and set
     # as attributes on the booster after it is loaded.
-    meta_data_attributes: Dict[str, BaseSerializer] = {
+    metadata_attributes: Dict[str, BaseSerializer] = {
         "feature_names_in_": ObjectSerializer(),
         "n_features_": ObjectSerializer(),
         "feature_importance_method": ObjectSerializer(),
@@ -38,6 +38,7 @@ class PerpetualBooster:
         self,
         *,
         objective: str = "LogLoss",
+        budget: float = 0.5,
         num_threads: Optional[int] = None,
         monotone_constraints: Union[Dict[Any, int], None] = None,
         force_children_to_bound_parent: bool = False,
@@ -48,8 +49,7 @@ class PerpetualBooster:
         missing_node_treatment: str = "None",
         log_iterations: int = 0,
         feature_importance_method: str = "Gain",
-        budget: Optional[float] = None,
-        alpha: Optional[float] = None,
+        quantile: Optional[float] = None,
         reset: Optional[bool] = None,
         categorical_features: Union[Iterable[int], Iterable[str], str, None] = "auto",
         timeout: Optional[float] = None,
@@ -59,16 +59,17 @@ class PerpetualBooster:
         max_bin: int = 256,
         max_cat: int = 1000,
     ):
-        """PerpetualBooster class, used to generate gradient boosted decision tree ensembles.
-        The following parameters can also be specified in the fit method to override the values in the constructor:
-            budget, alpha, reset, categorical_features, timeout, iteration_limit, memory_limit, and stopping_rounds.
+        """PerpetualBooster class, used to create gradient boosted decision tree ensembles.
 
         Args:
-            objective (str, optional): Learning objective function to be used for optimization.
-                Valid options include "LogLoss" to use logistic loss (classification),
+            objective (str, optional): Learning objective function to be used for optimization. Valid options are:
+                "LogLoss" to use logistic loss (classification),
                 "SquaredLoss" to use squared error (regression),
                 "QuantileLoss" to use quantile error (regression).
                 Defaults to "LogLoss".
+            budget (float, optional): a positive number for fitting budget. Increasing this number will more
+                likely result in more boosting rounds and more increased predictive power.
+                Default value is 0.5.
             num_threads (int, optional): Number of threads to be used during training.
             monotone_constraints (Dict[Any, int], optional): Constraints that are used to enforce a
                 specific relationship between the training features and the target variable. A dictionary
@@ -105,10 +106,7 @@ class PerpetualBooster:
                 - "AverageNodeWeight": Set the missing node to be equal to the weighted average weight of the left and the right nodes.
             log_iterations (int, optional): Setting to a value (N) other than zero will result in information being logged about ever N iterations, info can be interacted with directly with the python [`logging`](https://docs.python.org/3/howto/logging.html) module. For an example of how to utilize the logging information see the example [here](/#logging-output).
             feature_importance_method (str, optional): The feature importance method type that will be used to calculate the `feature_importances_` attribute on the booster.
-            budget (float, optional): a positive number for fitting budget. Increasing this number will more
-                likely result in more boosting rounds and more increased predictive power.
-                Default value is 1.0.
-            alpha (float, optional): only used in quantile regression.
+            quantile (float, optional): only used in quantile regression.
             reset (bool, optional): whether to reset the model or continue training.
             categorical_features (Union[Iterable[int], Iterable[str], str, None], optional): The names or indices for categorical features.
                 Defaults to `auto` for Polars or Pandas categorical data types.
@@ -166,6 +164,7 @@ class PerpetualBooster:
         )
 
         self.objective = objective
+        self.budget = budget
         self.num_threads = num_threads
         self.monotone_constraints = monotone_constraints_
         self.force_children_to_bound_parent = force_children_to_bound_parent
@@ -176,8 +175,7 @@ class PerpetualBooster:
         self.missing_node_treatment = missing_node_treatment
         self.log_iterations = log_iterations
         self.feature_importance_method = feature_importance_method
-        self.budget = budget
-        self.alpha = alpha
+        self.quantile = quantile
         self.reset = reset
         self.categorical_features = categorical_features
         self.timeout = timeout
@@ -189,6 +187,7 @@ class PerpetualBooster:
 
         booster = CratePerpetualBooster(
             objective=self.objective,
+            budget=self.budget,
             max_bin=self.max_bin,
             num_threads=self.num_threads,
             monotone_constraints=dict(),
@@ -199,23 +198,17 @@ class PerpetualBooster:
             terminate_missing_features=set(),
             missing_node_treatment=self.missing_node_treatment,
             log_iterations=self.log_iterations,
+            quantile=self.quantile,
+            reset=self.reset,
+            categorical_features=set(),
+            timeout=self.timeout,
+            iteration_limit=self.iteration_limit,
+            memory_limit=self.memory_limit,
+            stopping_rounds=self.stopping_rounds,
         )
         self.booster = cast(BoosterType, booster)
 
-    def fit(
-        self,
-        X,
-        y,
-        sample_weight=None,
-        budget: Optional[float] = None,
-        alpha: Optional[float] = None,
-        reset: Optional[bool] = None,
-        categorical_features: Union[Iterable[int], Iterable[str], str, None] = "auto",
-        timeout: Optional[float] = None,
-        iteration_limit: Optional[int] = None,
-        memory_limit: Optional[float] = None,
-        stopping_rounds: Optional[int] = None,
-    ) -> Self:
+    def fit(self, X, y, sample_weight=None) -> Self:
         """Fit the gradient booster on a provided dataset.
 
         Args:
@@ -225,26 +218,10 @@ class PerpetualBooster:
             sample_weight (Union[ArrayLike, None], optional): Instance weights to use when
                 training the model. If None is passed, a weight of 1 will be used for every record.
                 Defaults to None.
-            budget (float, optional): a positive number for fitting budget. Increasing this number will more
-                likely result in more boosting rounds and more increased predictive power.
-                Defaults to 1.0.
-            alpha (float, optional): only used in quantile regression.
-            reset (bool, optional): whether to reset the model or continue training.
-            categorical_features (Union[Iterable[int], Iterable[str], str, None], optional): The names or indices for categorical features.
-                Defaults to `auto` for Polars or Pandas categorical data types.
-            timeout (float, optional): optional fit timeout in seconds
-            iteration_limit (int, optional): optional limit for the number of boosting rounds. The default value is 1000 boosting rounds.
-                The algorithm automatically stops for most of the cases before hitting this limit.
-                If you want to experiment with very high budget (>2.0), you can also increase this limit.
-            memory_limit (float, optional): optional limit for memory allocation in GB. If not set, the memory will be allocated based on
-                available memory and the algorithm requirements.
-            stopping_rounds (int, optional): optional limit for auto stopping. Defaults to 3.
         """
 
         features_, flat_data, rows, cols, categorical_features_, cat_mapping = (
-            convert_input_frame(
-                X, categorical_features or self.categorical_features, self.max_cat
-            )
+            convert_input_frame(X, self.categorical_features, self.max_cat)
         )
         self.n_features_ = cols
         self.cat_mapping = cat_mapping
@@ -268,6 +245,7 @@ class PerpetualBooster:
         ):
             booster = CratePerpetualBooster(
                 objective=self.objective,
+                budget=self.budget,
                 max_bin=self.max_bin,
                 num_threads=self.num_threads,
                 monotone_constraints=crate_mc,
@@ -278,12 +256,20 @@ class PerpetualBooster:
                 terminate_missing_features=crate_tmf,
                 missing_node_treatment=self.missing_node_treatment,
                 log_iterations=self.log_iterations,
+                quantile=self.quantile,
+                reset=self.reset,
+                categorical_features=categorical_features_,
+                timeout=self.timeout,
+                iteration_limit=self.iteration_limit,
+                memory_limit=self.memory_limit,
+                stopping_rounds=self.stopping_rounds,
             )
             self.booster = cast(BoosterType, booster)
         else:
             booster = CrateMultiOutputBooster(
                 n_boosters=len(classes_),
                 objective=self.objective,
+                budget=self.budget,
                 max_bin=self.max_bin,
                 num_threads=self.num_threads,
                 monotone_constraints=crate_mc,
@@ -294,6 +280,13 @@ class PerpetualBooster:
                 terminate_missing_features=crate_tmf,
                 missing_node_treatment=self.missing_node_treatment,
                 log_iterations=self.log_iterations,
+                quantile=self.quantile,
+                reset=self.reset,
+                categorical_features=categorical_features_,
+                timeout=self.timeout,
+                iteration_limit=self.iteration_limit,
+                memory_limit=self.memory_limit,
+                stopping_rounds=self.stopping_rounds,
             )
             self.booster = cast(MultiOutputBoosterType, booster)
 
@@ -305,20 +298,97 @@ class PerpetualBooster:
         )
         self._set_metadata_attributes("classes_", self.classes_)
 
+        self.categorical_features = categorical_features_
+
         self.booster.fit(
             flat_data=flat_data,
             rows=rows,
             cols=cols,
             y=y_,
-            budget=budget or self.budget,
             sample_weight=sample_weight_,  # type: ignore
-            alpha=alpha or self.alpha,
-            reset=reset or self.reset,
-            categorical_features=categorical_features_,  # type: ignore
-            timeout=timeout or self.timeout,
-            iteration_limit=iteration_limit or self.iteration_limit,
-            memory_limit=memory_limit or self.memory_limit,
-            stopping_rounds=stopping_rounds or self.stopping_rounds,
+        )
+
+        return self
+
+    def prune(self, X, y, sample_weight=None) -> Self:
+        """Prune the gradient booster on a provided dataset.
+
+        Args:
+            X (FrameLike): Either a Polars or Pandas DataFrame, or a 2 dimensional Numpy array.
+            y (Union[FrameLike, ArrayLike]): Either a Polars or Pandas DataFrame or Series,
+                or a 1 or 2 dimensional Numpy array.
+            sample_weight (Union[ArrayLike, None], optional): Instance weights to use when
+                training the model. If None is passed, a weight of 1 will be used for every record.
+                Defaults to None.
+        """
+
+        _, flat_data, rows, cols = transform_input_frame(X, self.cat_mapping)
+
+        y_, _ = convert_input_array(y, self.objective)
+
+        if sample_weight is None:
+            sample_weight_ = None
+        else:
+            sample_weight_, _ = convert_input_array(sample_weight, self.objective)
+
+        self.booster.prune(
+            flat_data=flat_data,
+            rows=rows,
+            cols=cols,
+            y=y_,
+            sample_weight=sample_weight_,  # type: ignore
+        )
+
+        return self
+
+    def calibrate(
+        self, X_train, y_train, X_cal, y_cal, alpha, sample_weight=None
+    ) -> Self:
+        """Calibrate the gradient booster on a provided dataset.
+
+        Args:
+            X_train (FrameLike): Either a Polars or Pandas DataFrame, or a 2 dimensional Numpy array.
+            y_train (Union[FrameLike, ArrayLike]): Either a Polars or Pandas DataFrame or Series,
+                or a 1 or 2 dimensional Numpy array.
+            X_cal (FrameLike): Either a Polars or Pandas DataFrame, or a 2 dimensional Numpy array.
+            y_cal (Union[FrameLike, ArrayLike]): Either a Polars or Pandas DataFrame or Series,
+                or a 1 or 2 dimensional Numpy array.
+            alpha (ArrayLike): Between 0 and 1, represents the uncertainty of the confidence interval.
+                Lower alpha produce larger (more conservative) prediction intervals.
+                alpha is the complement of the target coverage level.
+            sample_weight (Union[ArrayLike, None], optional): Instance weights to use when
+                training the model. If None is passed, a weight of 1 will be used for every record.
+                Defaults to None.
+        """
+
+        _, flat_data_train, rows_train, cols_train = transform_input_frame(
+            X_train, self.cat_mapping
+        )
+
+        y_train_, _ = convert_input_array(y_train, self.objective)
+
+        _, flat_data_cal, rows_cal, cols_cal = transform_input_frame(
+            X_cal, self.cat_mapping
+        )
+
+        y_cal_, _ = convert_input_array(y_cal, self.objective)
+
+        if sample_weight is None:
+            sample_weight_ = None
+        else:
+            sample_weight_, _ = convert_input_array(sample_weight, self.objective)
+
+        self.booster.calibrate(
+            flat_data=flat_data_train,
+            rows=rows_train,
+            cols=cols_train,
+            y=y_train_,
+            flat_data_cal=flat_data_cal,
+            rows_cal=rows_cal,
+            cols_cal=cols_cal,
+            y_cal=y_cal_,
+            alpha=np.array(alpha),
+            sample_weight=sample_weight_,  # type: ignore
         )
 
         return self
@@ -331,6 +401,29 @@ class PerpetualBooster:
                         f"Columns mismatch between data {features} passed, and data {self.feature_names_in_} used at fit."
                     )
 
+    def predict_intervals(self, X, parallel: Union[bool, None] = None) -> dict:
+        """Predict intervals with the fitted booster on new data.
+
+        Args:
+            X (FrameLike): Either a Polars or Pandas DataFrame, or a 2 dimensional Numpy array.
+            parallel (Union[bool, None], optional): Optionally specify if the predict
+                function should run in parallel on multiple threads. If `None` is
+                passed, the `parallel` attribute of the booster will be used.
+                Defaults to `None`.
+
+        Returns:
+            np.ndarray: Returns a numpy array of the predictions.
+        """
+        features_, flat_data, rows, cols = transform_input_frame(X, self.cat_mapping)
+        self._validate_features(features_)
+
+        return self.booster.predict_intervals(
+            flat_data=flat_data,
+            rows=rows,
+            cols=cols,
+            parallel=parallel,
+        )
+
     def predict(self, X, parallel: Union[bool, None] = None) -> np.ndarray:
         """Predict with the fitted booster on new data.
 
@@ -699,14 +792,17 @@ class PerpetualBooster:
         Returns:
             PerpetualBooster: An initialized booster object.
         """
-        booster = CratePerpetualBooster.load_booster(str(path))
+        try:
+            booster = CratePerpetualBooster.load_booster(str(path))
+        except ValueError:
+            booster = CrateMultiOutputBooster.load_booster(str(path))
 
         params = booster.get_params()
         with warnings.catch_warnings():
             warnings.simplefilter("ignore")
             c = cls(**params)
         c.booster = booster
-        for m in c.meta_data_attributes:
+        for m in c.metadata_attributes:
             try:
                 m_ = c._get_metadata_attributes(m)
                 setattr(c, m, m_)
@@ -774,12 +870,12 @@ class PerpetualBooster:
         return v
 
     def _set_metadata_attributes(self, key: str, value: Any) -> None:
-        value_ = self.meta_data_attributes[key].serialize(value)
+        value_ = self.metadata_attributes[key].serialize(value)
         self.insert_metadata(key=key, value=value_)
 
     def _get_metadata_attributes(self, key: str) -> Any:
         value = self.get_metadata(key)
-        return self.meta_data_attributes[key].deserialize(value)
+        return self.metadata_attributes[key].deserialize(value)
 
     @property
     def base_score(self) -> Union[float, Iterable[float]]:
@@ -810,7 +906,12 @@ class PerpetualBooster:
 
     def __setstate__(self, d: Dict[Any, Any]) -> None:
         # Load the booster object the pickled JSon string.
-        booster_object = CratePerpetualBooster.from_json(d["__booster_json_file__"])
+        try:
+            booster_object = CratePerpetualBooster.from_json(d["__booster_json_file__"])
+        except ValueError:
+            booster_object = CrateMultiOutputBooster.from_json(
+                d["__booster_json_file__"]
+            )
         d["booster"] = booster_object
         # Are there any new parameters, that need to be added to the python object,
         # that would have been loaded in as defaults on the json object?

perpetual/utils.py

@@ -65,7 +65,9 @@ def convert_input_array(x, objective) -> np.ndarray:
 
 
 def convert_input_frame(
-    X, categorical_features, max_cat
+    X,
+    categorical_features,
+    max_cat,
 ) -> Tuple[List[str], np.ndarray, int, int, Optional[Iterable[int]], Optional[Dict]]:
     """Convert data to format needed by booster.
 

{perpetual-0.7.12.dist-info → perpetual-0.8.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: perpetual
-Version: 0.7.12
+Version: 0.8.1
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3.9
@@ -10,6 +10,7 @@ Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Dist: numpy
 Requires-Dist: typing-extensions
+Requires-Dist: black ; extra == 'dev'
 Requires-Dist: pandas ; extra == 'dev'
 Requires-Dist: polars ; extra == 'dev'
 Requires-Dist: pyarrow ; extra == 'dev'
@@ -24,7 +25,7 @@ Requires-Dist: ruff ; extra == 'dev'
 Provides-Extra: dev
 License-File: LICENSE
 License-File: LICENSE
-Summary: A self-generalizing gradient boosting machine which doesn't need hyperparameter optimization
+Summary: A self-generalizing gradient boosting machine that doesn't need hyperparameter optimization
 Keywords: rust,perpetual,machine learning,tree model,decision tree,gradient boosted decision tree,gradient boosting machine
 Home-Page: https://perpetual-ml.com
 Author: Mutlu Simsek
@@ -49,10 +50,28 @@ Project-URL: Source Code, https://github.com/perpetual-ml/perpetual
 
 # Perpetual
 
-PerpetualBooster is a gradient boosting machine (GBM) algorithm which doesn't need hyperparameter optimization unlike other GBM algorithms. Similar to AutoML libraries, it has a `budget` parameter. Increasing the `budget` parameter increases the predictive power of the algorithm and gives better results on unseen data. Start with a small budget (e.g. 1.0) and increase it (e.g. 2.0) once you are confident with your features. If you don't see any improvement with further increasing the `budget`, it means that you are already extracting the most predictive power out of your data.
+PerpetualBooster is a gradient boosting machine (GBM) algorithm that doesn't need hyperparameter optimization unlike other GBM algorithms. Similar to AutoML libraries, it has a `budget` parameter. Increasing the `budget` parameter increases the predictive power of the algorithm and gives better results on unseen data. Start with a small budget (e.g. 0.5) and increase it (e.g. 1.0) once you are confident with your features. If you don't see any improvement with further increasing the `budget`, it means that you are already extracting the most predictive power out of your data.
+
+## Usage
+
+You can use the algorithm like in the example below. Check examples folders for both Rust and Python.
+
+```python
+from perpetual import PerpetualBooster
+
+model = PerpetualBooster(objective="SquaredLoss", budget=0.5)
+model.fit(X, y)
+```
+
+## Documentation
+
+Documentation for the Python API can be found [here](https://perpetual-ml.github.io/perpetual) and for the Rust API [here](https://docs.rs/perpetual/latest/perpetual/).
+
 
 ## Benchmark
 
+### PerpetualBooster vs. Optuna + LightGBM
+
 Hyperparameter optimization usually takes 100 iterations with plain GBM algorithms. PerpetualBooster achieves the same accuracy in a single run. Thus, it achieves up to 100x speed-up at the same accuracy with different `budget` levels and with different datasets.
 
 The following table summarizes the results for the [California Housing](https://scikit-learn.org/stable/modules/generated/sklearn.datasets.fetch_california_housing.html) dataset (regression):
@@ -71,38 +90,51 @@ The following table summarizes the results for the [Cover Types](https://scikit-
 
 The results can be reproduced using the scripts in the [examples](./python-package/examples) folder.
 
-PerpetualBooster is a GBM but behaves like AutoML so it is benchmarked also against AutoGluon (v1.2, best quality preset), the current leader in [AutoML benchmark](https://automlbenchmark.streamlit.app/cd_diagram). Top 10 datasets with the most number of rows are selected from [OpenML datasets](https://www.openml.org/). The results are summarized in the following table for regression tasks:
+### PerpetualBooster vs. AutoGluon
 
-| OpenML Task                                  | Perpetual Training Duration | Perpetual Inference Duration                                      | Perpetual RMSE | AutoGluon Training Duration | AutoGluon Inference Duration                                      | AutoGluon RMSE |
-| -------------------------------------------- | --------------------------- | ----------------------------------------------------------------- | -------------- | --------------------------- | ----------------------------------------------------------------- | -------------- |
-| [Airlines_DepDelay_10M](https://www.openml.org/t/359929) | 518                         | 11.3                                                              | 29.0           | 520                         | 30.9 <td style="background-color:green;color:white;"> 28.8 </td>  |
-| [bates_regr_100](https://www.openml.org/t/361940)        | 3421                        | 15.1 <td style="background-color:green;color:white;"> 1.084 </td> | OOM            | OOM                         | OOM                                                               |
-| [BNG(libras_move)](https://www.openml.org/t/7327)        | 1956                        | 4.2 <td style="background-color:green;color:white;"> 2.51 </td>   | 1922           | 97.6                        | 2.53                                                              |
-| [BNG(satellite_image)](https://www.openml.org/t/7326)    | 334                         | 1.6                                                               | 0.731          | 337                         | 10.0 <td style="background-color:green;color:white;"> 0.721 </td> |
-| [COMET_MC](https://www.openml.org/t/14949)               | 44                          | 1.0 <td style="background-color:green;color:white;"> 0.0615 </td> | 47             | 5.0                         | 0.0662                                                            |
-| [friedman1](https://www.openml.org/t/361939)             | 275                         | 4.2 <td style="background-color:green;color:white;"> 1.047 </td>  | 278            | 5.1                         | 1.487                                                             |
-| [poker](https://www.openml.org/t/10102)                  | 38                          | 0.6 <td style="background-color:green;color:white;"> 0.256 </td>  | 41             | 1.2                         | 0.722                                                             |
-| [subset_higgs](https://www.openml.org/t/361955)          | 868                         | 10.6 <td style="background-color:green;color:white;"> 0.420 </td> | 870            | 24.5                        | 0.421                                                             |
-| [BNG(autoHorse)](https://www.openml.org/t/7319)          | 107                         | 1.1 <td style="background-color:green;color:white;"> 19.0 </td>   | 107            | 3.2                         | 20.5                                                              |
-| [BNG(pbc)](https://www.openml.org/t/7318)                | 48                          | 0.6 <td style="background-color:green;color:white;"> 836.5 </td>  | 51             | 0.2                         | 957.1                                                             |
-| average                                      | 465                         | 3.9                                                               | -              | 464                         | 19.7                                                              | -              |
+PerpetualBooster is a GBM but behaves like AutoML so it is benchmarked also against AutoGluon (v1.2, best quality preset), the current leader in [AutoML benchmark](https://automlbenchmark.streamlit.app/cd_diagram). Top 10 datasets with the most number of rows are selected from [OpenML datasets](https://www.openml.org/) for both regression and classification tasks. 
 
-PerpetualBooster outperformed AutoGluon on 8 out of 10 datasets, training equally fast and inferring 5x faster. The results can be reproduced using the automlbenchmark fork [here](https://github.com/deadsoul44/automlbenchmark).
+The results are summarized in the following table for regression tasks:
 
-## Usage
+| OpenML Task | Perpetual Training Duration | Perpetual Inference Duration | Perpetual RMSE | AutoGluon Training Duration | AutoGluon Inference Duration | AutoGluon RMSE |
+| -------------------------------------------------------- | ----- | ----- | ------------------- | -------- | ------ | ------------------ |
+| [Airlines_DepDelay_10M](https://www.openml.org/t/359929) | 518   | 11.3  | 29.0                | 520      | 30.9   | <ins> 28.8 </ins>  |
+| [bates_regr_100](https://www.openml.org/t/361940)        | 3421  | 15.1  | <ins> 1.084 </ins>  | OOM      | OOM    | OOM                |
+| [BNG(libras_move)](https://www.openml.org/t/7327)        | 1956  | 4.2   | <ins> 2.51 </ins>   | 1922     | 97.6   | 2.53               |
+| [BNG(satellite_image)](https://www.openml.org/t/7326)    | 334   | 1.6   | 0.731               | 337      | 10.0   | <ins> 0.721 </ins> |
+| [COMET_MC](https://www.openml.org/t/14949)               | 44    | 1.0   | <ins> 0.0615 </ins> | 47       | 5.0    | 0.0662             |
+| [friedman1](https://www.openml.org/t/361939)             | 275   | 4.2   | <ins> 1.047 </ins>  | 278      | 5.1    | 1.487              |
+| [poker](https://www.openml.org/t/10102)                  | 38    | 0.6   | <ins> 0.256 </ins>  | 41       | 1.2    | 0.722              |
+| [subset_higgs](https://www.openml.org/t/361955)          | 868   | 10.6  | <ins> 0.420 </ins>  | 870      | 24.5   | 0.421              |
+| [BNG(autoHorse)](https://www.openml.org/t/7319)          | 107   | 1.1   | <ins> 19.0 </ins>   | 107      | 3.2    | 20.5               |
+| [BNG(pbc)](https://www.openml.org/t/7318)                | 48    | 0.6   | <ins> 836.5 </ins>  | 51       | 0.2    | 957.1              |
+| average                                                  | 465   | 3.9   | -                   | 464      | 19.7   | -                  |
 
-You can use the algorithm like in the example below. Check examples folders for both Rust and Python.
+PerpetualBooster outperformed AutoGluon on 8 out of 10 regression tasks, training equally fast and inferring 5.1x faster. 
 
-```python
-from perpetual import PerpetualBooster
+The results are summarized in the following table for classification tasks:
 
-model = PerpetualBooster(objective="SquaredLoss")
-model.fit(X, y, budget=1.0)
-```
+| OpenML Task | Perpetual Training Duration | Perpetual Inference Duration | Perpetual AUC | AutoGluon Training Duration | AutoGluon Inference Duration | AutoGluon AUC |
+| -------------------------------------------------------- | ------- | ------ | ------------------- | -------- | ------ | ------------------ |
+| [BNG(spambase)](https://www.openml.org/t/146163)         | 70.1    | 2.1   | <ins> 0.671 </ins> | 73.1     | 3.7    | 0.669              |
+| [BNG(trains)](https://www.openml.org/t/208)              | 89.5    | 1.7   | <ins> 0.996 </ins> | 106.4    | 2.4    | 0.994              |
+| [breast](https://www.openml.org/t/361942)                | 13699.3 | 97.7  | <ins> 0.991 </ins> | 13330.7  | 79.7   | 0.949              |
+| [Click_prediction_small](https://www.openml.org/t/7291)  | 89.1    | 1.0   | <ins> 0.749 </ins> | 101.0    | 2.8    | 0.703              |
+| [colon](https://www.openml.org/t/361938)                 | 12435.2 | 126.7 | <ins> 0.997 </ins> | 12356.2  | 152.3  | 0.997              |
+| [Higgs](https://www.openml.org/t/362113)                 | 3485.3  | 40.9  | <ins> 0.843 </ins> | 3501.4   | 67.9   | 0.816              |
+| [SEA(50000)](https://www.openml.org/t/230)               | 21.9    | 0.2   | <ins> 0.936 </ins> | 25.6     | 0.5    | 0.935              |
+| [sf-police-incidents](https://www.openml.org/t/359994)   | 85.8    | 1.5   | <ins> 0.687 </ins> | 99.4     | 2.8    | 0.659              |
+| [bates_classif_100](https://www.openml.org/t/361941)     | 11152.8 | 50.0  | <ins> 0.864 </ins> | OOM      | OOM    | OOM                |
+| [prostate](https://www.openml.org/t/361945)              | 13699.9 | 79.8  | <ins> 0.987 </ins> | OOM      | OOM    | OOM                |
+| average                                                  | 3747.0  | 34.0  | -                  | 3699.2   | 39.0   | -                  |
+
+PerpetualBooster outperformed AutoGluon on 10 out of 10 classification tasks, training equally fast and inferring 1.1x faster. 
+
+PerpetualBooster demonstrates greater robustness compared to AutoGluon, successfully training on all 20 tasks, whereas AutoGluon encountered out-of-memory errors on 3 of those tasks.
+
+The results can be reproduced using the automlbenchmark fork [here](https://github.com/deadsoul44/automlbenchmark).
 
-## Documentation
 
-Documentation for the Python API can be found [here](https://perpetual-ml.github.io/perpetual) and for the Rust API [here](https://docs.rs/perpetual/latest/perpetual/).
 
 ## Installation
 

perpetual-0.8.1.dist-info/RECORD

@@ -0,0 +1,12 @@
+perpetual-0.8.1.dist-info/METADATA,sha256=UNcSudsW5App4W9EnFAgjxFrhbHwfbomtz7jwGIVi5s,10752
+perpetual-0.8.1.dist-info/WHEEL,sha256=iNzfSeughQ6gviCftXhu6zZQCMTOJAdqefPsfmeKgU8,95
+perpetual-0.8.1.dist-info/license_files/LICENSE,sha256=gcuuhKKc5-dwvyvHsXjlC9oM6N5gZ6umYbC8ewW1Yvg,35821
+perpetual-0.8.1.dist-info/license_files/LICENSE,sha256=gcuuhKKc5-dwvyvHsXjlC9oM6N5gZ6umYbC8ewW1Yvg,35821
+perpetual/booster.py,sha256=ne-RgsYIjzQAYiTtI1PJ_IpolnS-C89trIFCdoiZoH4,50118
+perpetual/data.py,sha256=HiDsv2i1p9cLkXe8vnekxfpafyuxfWXwXrucdIir3xk,614
+perpetual/serialize.py,sha256=FeW4JsUFVsrft9N7gz-ebn5mXvDv4LiJC2sgBEeGxYo,1957
+perpetual/types.py,sha256=idZNsDErNTur_rJ_5Co8Pb6fik-AUn9lkrXmjbQJVX0,3381
+perpetual/utils.py,sha256=nqwO6GFHi7I5iltuvgLT3NFaPm1h9cHlnomjFcdSfHY,7455
+perpetual/__init__.py,sha256=V0RhghaG0CuKxKrzYUBYqrf7Drb-gjmznsbz9KT12lk,122
+perpetual/perpetual.cp313-win_amd64.pyd,sha256=lrWyuEZiMH8l75Tprn383NMkPDGh7q18LRIlKYzzE14,1661952
+perpetual-0.8.1.dist-info/RECORD,,

perpetual-0.7.12.dist-info/RECORD

@@ -1,12 +0,0 @@
-perpetual-0.7.12.dist-info/METADATA,sha256=Sfq0haXk0OttukvSaG6ARYPQipfeQ7ZIO_m7iaqkvys,10014
-perpetual-0.7.12.dist-info/WHEEL,sha256=iNzfSeughQ6gviCftXhu6zZQCMTOJAdqefPsfmeKgU8,95
-perpetual-0.7.12.dist-info/license_files/LICENSE,sha256=gcuuhKKc5-dwvyvHsXjlC9oM6N5gZ6umYbC8ewW1Yvg,35821
-perpetual-0.7.12.dist-info/license_files/LICENSE,sha256=gcuuhKKc5-dwvyvHsXjlC9oM6N5gZ6umYbC8ewW1Yvg,35821
-perpetual/booster.py,sha256=ICWJRuSxoaUgRHo9N8hodz1MlyRBVKPhVnfQJOes968,46919
-perpetual/data.py,sha256=HiDsv2i1p9cLkXe8vnekxfpafyuxfWXwXrucdIir3xk,614
-perpetual/serialize.py,sha256=FeW4JsUFVsrft9N7gz-ebn5mXvDv4LiJC2sgBEeGxYo,1957
-perpetual/types.py,sha256=idZNsDErNTur_rJ_5Co8Pb6fik-AUn9lkrXmjbQJVX0,3381
-perpetual/utils.py,sha256=i_7EB5xQXAGtODONhrOwfxRfH3YR7U0cQJvL8eUNFK8,7444
-perpetual/__init__.py,sha256=V0RhghaG0CuKxKrzYUBYqrf7Drb-gjmznsbz9KT12lk,122
-perpetual/perpetual.cp313-win_amd64.pyd,sha256=wB4UC94u1mcKNicec_h62WjjJJwp1WOerPjMAcKyELY,1509376
-perpetual-0.7.12.dist-info/RECORD,,