feat: scikit-base registry and testing framework

.github/workflows/CI.yml

@@ -72,10 +72,7 @@ jobs:
         run: git branch -a
 
       - name: Run tests
-        run: python -m pytest tests -vv
-
-      - name: Publish code coverage
-        uses: codecov/codecov-action@v4
+        run: python -m pytest src/ tests/ -vv
 
   test-all-softdeps:
     needs: test-no-softdeps
@@ -107,7 +104,7 @@ jobs:
         run: git branch -a
 
       - name: Run tests
-        run: python -m pytest tests -vv
+        run: python -m pytest src/ tests/ -vv --cov=src/
 
       - name: Publish code coverage
         uses: codecov/codecov-action@v4

src/tsbootstrap/base_bootstrap.py

@@ -35,7 +35,11 @@ class BaseTimeSeriesBootstrap(BaseObject):
         If n_bootstraps is not greater than 0.
     """
 
-    _tags = {"object_type": "bootstrap"}
+    _tags = {
+        "object_type": "bootstrap",
+        "bootstrap_type": "other",
+        "capability:multivariate": True,
+    }
 
     def __init__(
         self,
@@ -117,7 +121,7 @@ def bootstrap(
         X_train, X_test = time_series_split(X, test_ratio=test_ratio)
 
         if y is not None:
-            self._check_input(y)
+            self._check_input(y, enforce_univariate=False)
             exog_train, _ = time_series_split(y, test_ratio=test_ratio)
         else:
             exog_train = None
@@ -150,8 +154,14 @@ def _generate_samples(
         for _ in range(self.config.n_bootstraps):
             indices, data = self._generate_samples_single_bootstrap(X=X, y=y)
             data = np.concatenate(data, axis=0)
+
+            # hack to fix known issue with non-concatenated index sets
+            # see bug issue #81
+            if isinstance(indices, list):
+                indices = np.concatenate(indices, axis=0)
+
             if return_indices:
-                yield indices, data  # type: ignore
+                yield data, indices  # type: ignore
             else:
                 yield data
 
@@ -162,11 +172,20 @@ def _generate_samples_single_bootstrap(self, X: np.ndarray, y=None):
         """
         raise NotImplementedError("abstract method")
 
-    def _check_input(self, X):
+    def _check_input(self, X, enforce_univariate=True):
         """Checks if the input is valid."""
         if np.any(np.diff([len(x) for x in X]) != 0):
             raise ValueError("All time series must be of the same length.")
 
+        self_can_only_univariate = not self.get_tag("capability:multivariate")
+        check_univariate = enforce_univariate and self_can_only_univariate
+        if check_univariate and X.shape[1] > 1:
+            raise ValueError(
+                f"Unsupported input type: the estimator {type(self)} "
+                "does not support multivariate endogeneous time series (X argument). "
+                "Pass an 1D np.array, or a 2D np.array with a single column."
+            )
+
     def get_n_bootstraps(
         self,
         X=None,
@@ -220,6 +239,12 @@ class BaseResidualBootstrap(BaseTimeSeriesBootstrap):
     _fit_model : Fits the model to the data and stores the residuals.
     """
 
+    _tags = {
+        "python_dependencies": "statsmodels",
+        "bootstrap_type": "residual",
+        "capability:multivariate": False,
+    }
+
     def __init__(
         self,
         n_bootstraps: Integral = 10,  # type: ignore
@@ -299,11 +324,14 @@ def _fit_model(self, X: np.ndarray, y=None) -> None:
             or self.fit_model is None
             or self.coefs is None
         ):
+            model_params = self.config.model_params
+            if model_params is None:
+                model_params = {}
             fit_obj = TSFitBestLag(
                 model_type=self.config.model_type,
                 order=self.config.order,
                 save_models=self.config.save_models,
-                **self.config.model_params,
+                **model_params,
             )
             self.fit_model = fit_obj.fit(X=X, y=y).model
             self.X_fitted = fit_obj.get_fitted_X()
@@ -486,7 +514,7 @@ class BaseStatisticPreservingBootstrap(BaseTimeSeriesBootstrap):
     def __init__(
         self,
         n_bootstraps: Integral = 10,  # type: ignore
-        statistic: Callable = np.mean,
+        statistic: Callable = None,
         statistic_axis: Integral = 0,  # type: ignore
         statistic_keepdims: bool = False,
         rng=None,
@@ -505,6 +533,9 @@ def __init__(
         self.statistic_axis = statistic_axis
         self.statistic_keepdims = statistic_keepdims
 
+        if statistic is None:
+            statistic = np.mean
+
         self.config = BaseStatisticPreservingBootstrapConfig(
             n_bootstraps=n_bootstraps,
             rng=rng,

src/tsbootstrap/base_bootstrap_configs.py

@@ -44,6 +44,8 @@ class BaseTimeSeriesBootstrapConfig(BaseObject):
     that are common to all time series bootstrapping methods.
     """
 
+    _tags = {"object_type": "config"}
+
     def __init__(
         self,
         n_bootstraps: Integral = 10,  # type: ignore
@@ -99,6 +101,7 @@ def __init__(
         model_type: ModelTypesWithoutArch = "ar",
         order=None,
         save_models: bool = False,
+        model_params=None,
         **kwargs,
     ):
         """
@@ -148,7 +151,7 @@ def __init__(
         self.model_type = model_type
         self.order = order
         self.save_models = save_models
-        self.model_params = kwargs
+        self.model_params = model_params
 
         super().__init__(n_bootstraps=n_bootstraps, rng=rng)
 

src/tsbootstrap/block_bootstrap.py

@@ -60,6 +60,8 @@ class BlockBootstrap(BaseTimeSeriesBootstrap):
         If block_length is not greater than 0.
     """
 
+    _tags = {"bootstrap_type": "block"}
+
     def __init__(
         self,
         n_bootstraps: Integral = 10,  # type: ignore
@@ -109,8 +111,8 @@ def __init__(
         self.blocks = None
         self.block_resampler = None
 
-    def _check_input(self, X: np.ndarray) -> None:
-        super()._check_input(X)
+    def _check_input(self, X: np.ndarray, enforce_univariate=True) -> None:
+        super()._check_input(X=X, enforce_univariate=enforce_univariate)
         if self.config.block_length is not None and self.config.block_length > X.shape[0]:  # type: ignore
             raise ValueError(
                 "block_length cannot be greater than the size of the input array X."

src/tsbootstrap/block_bootstrap_configs.py

@@ -303,7 +303,7 @@ class MovingBlockBootstrapConfig(BlockBootstrapConfig):
 
     def __init__(
         self,
-        block_length: Integral,
+        block_length: Integral = None,
         **kwargs,
     ) -> None:
         """
@@ -383,7 +383,7 @@ class CircularBlockBootstrapConfig(BlockBootstrapConfig):
 
     def __init__(
         self,
-        block_length: Integral,
+        block_length: Integral = None,
         **kwargs,
     ) -> None:
         """

src/tsbootstrap/block_length_sampler.py

@@ -53,6 +53,8 @@ class BlockLengthSampler(BaseObject):
         Sample a block length from the selected distribution.
     """
 
+    _tags = {"object_type": "sampler"}
+
     def __init__(self, avg_block_length: Integral = DEFAULT_AVG_BLOCK_LENGTH, block_length_distribution: str = None, rng: RngTypes = None):  # type: ignore
         """
         Initialize the BlockLengthSampler with the selected distribution and average block length.
@@ -76,6 +78,8 @@ def __init__(self, avg_block_length: Integral = DEFAULT_AVG_BLOCK_LENGTH, block_
         self.avg_block_length = avg_block_length
         self.rng = rng
 
+        super().__init__()
+
     @property
     def block_length_distribution(self) -> str:
         """Getter for block_length_distribution."""

src/tsbootstrap/bootstrap.py

@@ -182,7 +182,8 @@ def _generate_samples_single_bootstrap(
         bootstrap_samples = self.X_fitted + np.concatenate(block_data, axis=0)
         return block_indices, [bootstrap_samples]
 
-    def get_test_params(self):
+    @classmethod
+    def get_test_params(cls, parameter_set="default"):
         from tsbootstrap.block_bootstrap import MovingBlockBootstrap
         bs = MovingBlockBootstrap()
         return {"block_bootstrap": bs}
@@ -416,7 +417,8 @@ def _generate_samples_single_bootstrap(
 
         return block_indices, [bootstrap_samples]
 
-    def get_test_params(self):
+    @classmethod
+    def get_test_params(cls, parameter_set="default"):
         from tsbootstrap.block_bootstrap import MovingBlockBootstrap
         bs = MovingBlockBootstrap()
         return {"block_bootstrap": bs}
@@ -500,7 +502,7 @@ def __init__(
         self,
         block_bootstrap,
         n_bootstraps: Integral = 10,  # type: ignore
-        statistic=np.mean,
+        statistic=None,
         statistic_axis: Integral = 0,  # type: ignore
         statistic_keepdims: bool = False,
         rng=None,
@@ -543,7 +545,8 @@ def _generate_samples_single_bootstrap(
         bootstrap_samples = block_data_concat + bias
         return block_indices, [bootstrap_samples]
 
-    def get_test_params(self):
+    @classmethod
+    def get_test_params(cls, parameter_set="default"):
         from tsbootstrap.block_bootstrap import MovingBlockBootstrap
         bs = MovingBlockBootstrap()
         return {"block_bootstrap": bs}
@@ -763,7 +766,8 @@ def _generate_samples_single_bootstrap(
             bootstrap_samples = self.X_fitted + bootstrap_residuals
             return block_indices, [bootstrap_samples]
 
-    def get_test_params(self):
+    @classmethod
+    def get_test_params(cls, parameter_set="default"):
         from tsbootstrap.block_bootstrap import MovingBlockBootstrap
         bs = MovingBlockBootstrap()
         return {"block_bootstrap": bs}
@@ -947,7 +951,8 @@ def _generate_samples_single_bootstrap(
 
         return block_indices, [bootstrapped_samples]
 
-    def get_test_params(self):
+    @classmethod
+    def get_test_params(cls, parameter_set="default"):
         from tsbootstrap.block_bootstrap import MovingBlockBootstrap
         bs = MovingBlockBootstrap()
         return {"block_bootstrap": bs}

src/tsbootstrap/registry/__init__.py

@@ -0,0 +1,13 @@
+"""Registry and lookup functionality."""
+
+from tsbootstrap.registry._lookup import all_objects
+from tsbootstrap.registry._tags import (
+    OBJECT_TAG_LIST,
+    OBJECT_TAG_REGISTER,
+)
+
+__all__ = [
+    "OBJECT_TAG_LIST",
+    "OBJECT_TAG_REGISTER",
+    "all_objects",
+]