wavetrainer 0.0.24__tar.gz → 0.0.25__tar.gz

{wavetrainer-0.0.24/wavetrainer.egg-info → wavetrainer-0.0.25}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: wavetrainer
-Version: 0.0.24
+Version: 0.0.25
 Summary: A library for automatically finding the optimal model within feature and hyperparameter space.
 Home-page: https://github.com/8W9aG/wavetrainer
 Author: Will Sackfield
@@ -23,6 +23,11 @@ Requires-Dist: venn-abers>=1.4.6
 Requires-Dist: mapie>=0.9.2
 Requires-Dist: pytz>=2025.1
 Requires-Dist: torch>=2.6.0
+Requires-Dist: tabpfn>=2.0.6
+Requires-Dist: tabpfn-extensions>=0.0.4
+Requires-Dist: shap>=0.47.2
+Requires-Dist: hyperopt>=0.2.7
+Requires-Dist: pytest-is-running>=1.5.1
 
 # wavetrainer
 
@@ -52,6 +57,11 @@ Python 3.11.6:
 - [mapie](https://mapie.readthedocs.io/en/stable/)
 - [pytz](https://pythonhosted.org/pytz/)
 - [torch](https://pytorch.org/)
+- [tabpfn](https://github.com/PriorLabs/TabPFN)
+- [tabpfn-extensions](https://github.com/PriorLabs/tabpfn-extensions)
+- [shap](https://shap.readthedocs.io/en/latest/)
+- [hyperopt](https://hyperopt.github.io/hyperopt/)
+- [pytest-is-running](https://github.com/adamchainz/pytest-is-running)
 
 ## Raison D'être :thought_balloon:
 

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/README.md

@@ -26,6 +26,11 @@ Python 3.11.6:
 - [mapie](https://mapie.readthedocs.io/en/stable/)
 - [pytz](https://pythonhosted.org/pytz/)
 - [torch](https://pytorch.org/)
+- [tabpfn](https://github.com/PriorLabs/TabPFN)
+- [tabpfn-extensions](https://github.com/PriorLabs/tabpfn-extensions)
+- [shap](https://shap.readthedocs.io/en/latest/)
+- [hyperopt](https://hyperopt.github.io/hyperopt/)
+- [pytest-is-running](https://github.com/adamchainz/pytest-is-running)
 
 ## Raison D'être :thought_balloon:
 

wavetrainer-0.0.24/wavetrainer.egg-info/requires.txt → wavetrainer-0.0.25/requirements.txt

@@ -10,3 +10,8 @@ venn-abers>=1.4.6
 mapie>=0.9.2
 pytz>=2025.1
 torch>=2.6.0
+tabpfn>=2.0.6
+tabpfn-extensions>=0.0.4
+shap>=0.47.2
+hyperopt>=0.2.7
+pytest-is-running>=1.5.1

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/setup.py

@@ -23,7 +23,7 @@ def install_requires() -> typing.List[str]:
 
 setup(
     name='wavetrainer',
-    version='0.0.24',
+    version='0.0.25',
     description='A library for automatically finding the optimal model within feature and hyperparameter space.',
     long_description=long_description,
     long_description_content_type='text/markdown',

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/tests/trainer_test.py

@@ -14,7 +14,7 @@ class TestTrainer(unittest.TestCase):
     def test_trainer(self):
         with tempfile.TemporaryDirectory() as tmpdir:
             trainer = Trainer(tmpdir, walkforward_timedelta=datetime.timedelta(days=7), trials=1)
-            x_data = [i for i in range(100)]
+            x_data = [i for i in range(101)]
             x_index = [datetime.datetime(2022, 1, 1) + datetime.timedelta(days=i) for i in range(len(x_data))]
             df = pd.DataFrame(
                 data={

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/__init__.py

@@ -2,5 +2,5 @@
 
 from .create import create
 
-__VERSION__ = "0.0.24"
+__VERSION__ = "0.0.25"
 __all__ = ("create",)

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/model/catboost_model.py

@@ -61,12 +61,17 @@ class CatboostModel(Model):
     def estimator(self) -> Any:
         return self._provide_catboost()
 
+    @property
+    def supports_importances(self) -> bool:
+        return True
+
     def pre_fit(
         self,
         df: pd.DataFrame,
         y: pd.Series | pd.DataFrame | None,
         eval_x: pd.DataFrame | None = None,
         eval_y: pd.Series | pd.DataFrame | None = None,
+        w: pd.Series | None = None,
     ):
         if y is None:
             raise ValueError("y is null.")
@@ -75,6 +80,7 @@ class CatboostModel(Model):
             EVAL_SET_ARG_KEY: (eval_x, eval_y),
             CAT_FEATURES_ARG_KEY: df.select_dtypes(include="category").columns.tolist(),
             ORIGINAL_X_ARG_KEY: df,
+            "sample_weight": w,
         }
 
     def set_options(self, trial: optuna.Trial | optuna.trial.FrozenTrial) -> None:

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/model/model.py

@@ -1,5 +1,6 @@
 """The prototype model class."""
 
+# pylint: disable=too-many-arguments,too-many-positional-arguments
 from typing import Any
 
 import pandas as pd
@@ -24,12 +25,20 @@ class Model(Params, Fit):
         """The estimator backing the model."""
         raise NotImplementedError("estimator not implemented in parent class.")
 
+    @property
+    def supports_importances(self) -> bool:
+        """Whether this model supports feature importances."""
+        raise NotImplementedError(
+            "supports_importances not implemented in parent class."
+        )
+
     def pre_fit(
         self,
         df: pd.DataFrame,
         y: pd.Series | pd.DataFrame | None,
         eval_x: pd.DataFrame | None = None,
         eval_y: pd.Series | pd.DataFrame | None = None,
+        w: pd.Series | None = None,
     ) -> dict[str, Any]:
         """A call to make sure the model is prepared for the target type."""
         raise NotImplementedError("pre_fit not implemented in parent class.")

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/model/model_router.py

@@ -9,11 +9,13 @@ import pandas as pd
 
 from .catboost_model import CatboostModel
 from .model import Model
+from .tabpfn_model import TabPFNModel
 
 _MODEL_ROUTER_FILE = "model_router.json"
 _MODEL_KEY = "model"
 _MODELS = {
     CatboostModel.name(): CatboostModel,
+    TabPFNModel.name(): TabPFNModel,
 }
 
 
@@ -39,23 +41,30 @@ class ModelRouter(Model):
             raise ValueError("model is null")
         return model.estimator
 
+    @property
+    def supports_importances(self) -> bool:
+        model = self._model
+        if model is None:
+            raise ValueError("model is null")
+        return model.supports_importances
+
     def pre_fit(
         self,
         df: pd.DataFrame,
         y: pd.Series | pd.DataFrame | None,
         eval_x: pd.DataFrame | None = None,
         eval_y: pd.Series | pd.DataFrame | None = None,
+        w: pd.Series | None = None,
     ) -> dict[str, Any]:
         model = self._model
         if model is None:
             raise ValueError("model is null")
-        return model.pre_fit(df, y=y, eval_x=eval_x, eval_y=eval_y)
+        return model.pre_fit(df, y=y, eval_x=eval_x, eval_y=eval_y, w=w)
 
     def set_options(self, trial: optuna.Trial | optuna.trial.FrozenTrial) -> None:
-        self._model = _MODELS[
-            trial.suggest_categorical("model", list(_MODELS.keys()))
-        ]()
-        self._model.set_options(trial)
+        model = _MODELS[trial.suggest_categorical("model", list(_MODELS.keys()))]()
+        model.set_options(trial)
+        self._model = model
 
     def load(self, folder: str) -> None:
         with open(os.path.join(folder, _MODEL_ROUTER_FILE), encoding="utf8") as handle:

wavetrainer-0.0.25/wavetrainer/model/tabpfn_model.py

@@ -0,0 +1,145 @@
+"""A model that wraps tabpfn."""
+# pylint: disable=duplicate-code,too-many-arguments,too-many-positional-arguments
+
+import json
+import os
+import pickle
+from typing import Any, Self
+
+import optuna
+import pandas as pd
+import pytest_is_running
+import torch
+from tabpfn_extensions.post_hoc_ensembles.sklearn_interface import (  # type: ignore
+    AutoTabPFNClassifier, AutoTabPFNRegressor)
+
+from ..model_type import ModelType, determine_model_type
+from .model import PREDICTION_COLUMN, PROBABILITY_COLUMN_PREFIX, Model
+
+_MODEL_FILENAME = "model.pkl"
+_MODEL_PARAMS_FILENAME = "model_params.json"
+_MODEL_TYPE_KEY = "model_type"
+
+
+class TabPFNModel(Model):
+    """A class that uses TabPFN as a model."""
+
+    _tabpfn: AutoTabPFNClassifier | AutoTabPFNRegressor | None
+    _model_type: None | ModelType
+
+    @classmethod
+    def name(cls) -> str:
+        return "tabpfn"
+
+    def __init__(self) -> None:
+        super().__init__()
+        self._tabpfn = None
+        self._model_type = None
+
+    @property
+    def estimator(self) -> Any:
+        return self._provide_tabpfn()
+
+    @property
+    def supports_importances(self) -> bool:
+        return False
+
+    def pre_fit(
+        self,
+        df: pd.DataFrame,
+        y: pd.Series | pd.DataFrame | None,
+        eval_x: pd.DataFrame | None = None,
+        eval_y: pd.Series | pd.DataFrame | None = None,
+        w: pd.Series | None = None,
+    ):
+        if y is None:
+            raise ValueError("y is null.")
+        self._model_type = determine_model_type(y)
+        return {}
+
+    def set_options(self, trial: optuna.Trial | optuna.trial.FrozenTrial) -> None:
+        pass
+
+    def load(self, folder: str) -> None:
+        with open(os.path.join(folder, _MODEL_FILENAME), "rb") as f:
+            self._tabpfn = pickle.load(f)
+        with open(
+            os.path.join(folder, _MODEL_PARAMS_FILENAME), encoding="utf8"
+        ) as handle:
+            params = json.load(handle)
+            self._model_type = ModelType(params[_MODEL_TYPE_KEY])
+
+    def save(self, folder: str) -> None:
+        with open(os.path.join(folder, _MODEL_FILENAME), "wb") as f:
+            pickle.dump(self._tabpfn, f)
+        with open(
+            os.path.join(folder, _MODEL_PARAMS_FILENAME), "w", encoding="utf8"
+        ) as handle:
+            json.dump(
+                {
+                    _MODEL_TYPE_KEY: str(self._model_type),
+                },
+                handle,
+            )
+
+    def fit(
+        self,
+        df: pd.DataFrame,
+        y: pd.Series | pd.DataFrame | None = None,
+        w: pd.Series | None = None,
+        eval_x: pd.DataFrame | None = None,
+        eval_y: pd.Series | pd.DataFrame | None = None,
+    ) -> Self:
+        if y is None:
+            raise ValueError("y is null.")
+        self._model_type = determine_model_type(y)
+        tabpfn = self._provide_tabpfn()
+        tabpfn.fit(df, y)
+        return self
+
+    def transform(self, df: pd.DataFrame) -> pd.DataFrame:
+        tabpfn = self._provide_tabpfn()
+        if tabpfn is None:
+            raise ValueError("tabpfn is null")
+        pred = tabpfn.predict(df)
+        new_df = pd.DataFrame(
+            index=df.index,
+            data={
+                PREDICTION_COLUMN: pred.flatten(),
+            },
+        )
+        if isinstance(tabpfn, AutoTabPFNClassifier):
+            proba = tabpfn.predict_proba(df)
+            for i in range(proba.shape[1]):
+                new_df[f"{PROBABILITY_COLUMN_PREFIX}{i}"] = proba[:, i]
+        return new_df
+
+    def _provide_tabpfn(self) -> AutoTabPFNClassifier | AutoTabPFNRegressor:
+        tabpfn = self._tabpfn
+        if tabpfn is None:
+            max_time = 1 if pytest_is_running.is_running() else 120
+            match self._model_type:
+                case ModelType.BINARY:
+                    tabpfn = AutoTabPFNClassifier(
+                        max_time=max_time,
+                        device="cuda" if torch.cuda.is_available() else "cpu",
+                    )
+                case ModelType.REGRESSION:
+                    tabpfn = AutoTabPFNRegressor(
+                        max_time=max_time,
+                        device="cuda" if torch.cuda.is_available() else "cpu",
+                    )
+                case ModelType.BINNED_BINARY:
+                    tabpfn = AutoTabPFNClassifier(
+                        max_time=max_time,
+                        device="cuda" if torch.cuda.is_available() else "cpu",
+                    )
+                case ModelType.MULTI_CLASSIFICATION:
+                    tabpfn = AutoTabPFNClassifier(
+                        max_time=max_time,
+                        device="cuda" if torch.cuda.is_available() else "cpu",
+                    )
+            self._tabpfn = tabpfn
+        if tabpfn is None:
+            raise ValueError("tabpfn is null")
+        return tabpfn

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/reducer/combined_reducer.py

@@ -12,6 +12,7 @@ from .constant_reducer import ConstantReducer
 from .correlation_reducer import CorrelationReducer
 from .duplicate_reducer import DuplicateReducer
 from .nonnumeric_reducer import NonNumericReducer
+from .pca_reducer import PCAReducer
 from .reducer import Reducer
 from .unseen_reducer import UnseenReducer
 
@@ -32,6 +33,7 @@ class CombinedReducer(Reducer):
             ConstantReducer(),
             DuplicateReducer(),
             CorrelationReducer(),
+            PCAReducer(),
         ]
 
     @classmethod
@@ -59,6 +61,8 @@ class CombinedReducer(Reducer):
                     self._reducers.append(NonNumericReducer())
                 elif reducer_name == UnseenReducer.name():
                     self._reducers.append(UnseenReducer())
+                elif reducer_name == PCAReducer.name():
+                    self._reducers.append(PCAReducer())
         for reducer in self._reducers:
             reducer.load(folder)
 

wavetrainer-0.0.25/wavetrainer/reducer/pca_reducer.py

@@ -0,0 +1,60 @@
+"""A reducer that removes low variance columns."""
+
+import os
+from typing import Self
+
+import joblib  # type: ignore
+import optuna
+import pandas as pd
+from sklearn.decomposition import PCA  # type: ignore
+from sklearn.preprocessing import StandardScaler  # type: ignore
+
+from .reducer import Reducer
+
+_PCA_FILE = "pca.joblib"
+_PCA_SCALER_FILE = "pca_scaler.joblib"
+
+
+class PCAReducer(Reducer):
+    """A class that removes low variance columns from a dataframe."""
+
+    # pylint: disable=too-many-positional-arguments,too-many-arguments
+
+    def __init__(self):
+        super().__init__()
+        self._scaler = StandardScaler()
+        self._pca = PCA(n_components=300)
+
+    @classmethod
+    def name(cls) -> str:
+        return "pca"
+
+    def set_options(self, trial: optuna.Trial | optuna.trial.FrozenTrial) -> None:
+        pass
+
+    def load(self, folder: str) -> None:
+        self._scaler = joblib.load(os.path.join(folder, _PCA_SCALER_FILE))
+        self._pca = joblib.load(os.path.join(folder, _PCA_FILE))
+
+    def save(self, folder: str) -> None:
+        joblib.dump(self._scaler, os.path.join(folder, _PCA_SCALER_FILE))
+        joblib.dump(self._pca, os.path.join(folder, _PCA_FILE))
+
+    def fit(
+        self,
+        df: pd.DataFrame,
+        y: pd.Series | pd.DataFrame | None = None,
+        w: pd.Series | None = None,
+        eval_x: pd.DataFrame | None = None,
+        eval_y: pd.Series | pd.DataFrame | None = None,
+    ) -> Self:
+        if len(df.columns.values) < self._pca.n_components:  # type: ignore
+            return self
+        x_scaled = self._scaler.fit_transform(df)
+        self._pca.fit(x_scaled)
+        return self
+
+    def transform(self, df: pd.DataFrame) -> pd.DataFrame:
+        if len(df.columns.values) < self._pca.n_components:  # type: ignore
+            return df
+        return self._pca.transform(df)

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/selector/selector.py

@@ -49,8 +49,10 @@ class Selector(Params, Fit):
         eval_x: pd.DataFrame | None = None,
         eval_y: pd.Series | pd.DataFrame | None = None,
     ) -> Self:
+        if not self._model.supports_importances:
+            return self
         sklearn.set_config(enable_metadata_routing=False)
-        model_kwargs = self._model.pre_fit(df, y=y, eval_x=eval_x, eval_y=eval_y)
+        model_kwargs = self._model.pre_fit(df, y=y, eval_x=eval_x, eval_y=eval_y, w=w)
         if not isinstance(y, pd.Series):
             raise ValueError("y is not a series.")
         if len(df.columns) <= 1:
@@ -65,7 +67,7 @@ class Selector(Params, Fit):
             ),
         )
         try:
-            self._selector.fit(df, y=y, sample_weight=w, **model_kwargs)
+            self._selector.fit(df, y=y, **model_kwargs)
         except ValueError as exc:
             # Catch issues with 1 feature as a reduction target.
             logging.warning(str(exc))
@@ -76,7 +78,8 @@ class Selector(Params, Fit):
             return df
         selector = self._selector
         if selector is None:
-            raise ValueError("selector is null.")
+            logging.warning("selector is null")
+            return df
         try:
             return df[selector.get_feature_names_out()]
         except AttributeError as exc:

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer/trainer.py

@@ -266,7 +266,6 @@ class Trainer(Fit):
                         return float(r2_score(y_test, y_pred[[PREDICTION_COLUMN]]))
                     return float(f1_score(y_test, y_pred[[PREDICTION_COLUMN]]))
                 except WavetrainException as exc:
-                    logging.warning("WE DID NOT END UP TRAINING ANYTHING!!!!!")
                     logging.warning(str(exc))
                     return -1.0
 
@@ -343,6 +342,8 @@ class Trainer(Fit):
 
                 test_df = df.iloc[: train_len + count + test_len]
                 test_series = y_series.iloc[: train_len + count + test_len]
+                if len(test_df) <= 2:
+                    continue
 
                 if test_idx < start_validation_index:
 

{wavetrainer-0.0.24 → wavetrainer-0.0.25/wavetrainer.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: wavetrainer
-Version: 0.0.24
+Version: 0.0.25
 Summary: A library for automatically finding the optimal model within feature and hyperparameter space.
 Home-page: https://github.com/8W9aG/wavetrainer
 Author: Will Sackfield
@@ -23,6 +23,11 @@ Requires-Dist: venn-abers>=1.4.6
 Requires-Dist: mapie>=0.9.2
 Requires-Dist: pytz>=2025.1
 Requires-Dist: torch>=2.6.0
+Requires-Dist: tabpfn>=2.0.6
+Requires-Dist: tabpfn-extensions>=0.0.4
+Requires-Dist: shap>=0.47.2
+Requires-Dist: hyperopt>=0.2.7
+Requires-Dist: pytest-is-running>=1.5.1
 
 # wavetrainer
 
@@ -52,6 +57,11 @@ Python 3.11.6:
 - [mapie](https://mapie.readthedocs.io/en/stable/)
 - [pytz](https://pythonhosted.org/pytz/)
 - [torch](https://pytorch.org/)
+- [tabpfn](https://github.com/PriorLabs/TabPFN)
+- [tabpfn-extensions](https://github.com/PriorLabs/tabpfn-extensions)
+- [shap](https://shap.readthedocs.io/en/latest/)
+- [hyperopt](https://hyperopt.github.io/hyperopt/)
+- [pytest-is-running](https://github.com/adamchainz/pytest-is-running)
 
 ## Raison D'être :thought_balloon:
 

{wavetrainer-0.0.24 → wavetrainer-0.0.25}/wavetrainer.egg-info/SOURCES.txt

@@ -32,6 +32,7 @@ wavetrainer/model/catboost_model.py
 wavetrainer/model/catboost_regressor_wrap.py
 wavetrainer/model/model.py
 wavetrainer/model/model_router.py
+wavetrainer/model/tabpfn_model.py
 wavetrainer/reducer/__init__.py
 wavetrainer/reducer/base_selector_reducer.py
 wavetrainer/reducer/combined_reducer.py
@@ -39,6 +40,7 @@ wavetrainer/reducer/constant_reducer.py
 wavetrainer/reducer/correlation_reducer.py
 wavetrainer/reducer/duplicate_reducer.py
 wavetrainer/reducer/nonnumeric_reducer.py
+wavetrainer/reducer/pca_reducer.py
 wavetrainer/reducer/reducer.py
 wavetrainer/reducer/unseen_reducer.py
 wavetrainer/selector/__init__.py

wavetrainer-0.0.24/requirements.txt → wavetrainer-0.0.25/wavetrainer.egg-info/requires.txt

@@ -9,4 +9,9 @@ catboost>=1.2.7
 venn-abers>=1.4.6
 mapie>=0.9.2
 pytz>=2025.1
-torch>=2.6.0
+torch>=2.6.0
+tabpfn>=2.0.6
+tabpfn-extensions>=0.0.4
+shap>=0.47.2
+hyperopt>=0.2.7
+pytest-is-running>=1.5.1