deskit 1.3__tar.gz → 1.3.2__tar.gz

{deskit-1.3/src/deskit.egg-info → deskit-1.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deskit
-Version: 1.3
+Version: 1.3.2
 Summary: A Python library for Dynamic Ensemble Selection
 Author: Tikhon Vodyanov
 License-Expression: MIT
@@ -251,6 +251,8 @@ router = DEWSU(task="regression", metric=pinball, mode="min", k=20)
 
 Built-in metric strings: `accuracy`, `mae`, `mse`, `rmse`, `log_loss`, `prob_correct`.
 
+Alternatively, pass `metric=None` and populate the dictionary passed to fit() with your own per-sample computed scores instead of predictions.
+
 ---
 
 ## Data types

{deskit-1.3 → deskit-1.3.2}/README.md

@@ -220,6 +220,8 @@ router = DEWSU(task="regression", metric=pinball, mode="min", k=20)
 
 Built-in metric strings: `accuracy`, `mae`, `mse`, `rmse`, `log_loss`, `prob_correct`.
 
+Alternatively, pass `metric=None` and populate the dictionary passed to fit() with your own per-sample computed scores instead of predictions.
+
 ---
 
 ## Data types

{deskit-1.3 → deskit-1.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "deskit"
-version = "1.3"
+version = "1.3.2"
 description = "A Python library for Dynamic Ensemble Selection"
 readme = "README.md"
 license = "MIT"

{deskit-1.3 → deskit-1.3.2}/src/deskit/base/knnbase.py

@@ -49,12 +49,21 @@ class KNNBase(PredictBase, BaseRouter):
         This method expects pre-validated numpy arrays.
         """
         self.models = list(preds_dict.keys())
-        n_val       = len(y)
-        n_models    = len(self.models)
+        n_val = len(y)
+        n_models = len(self.models)
         self.matrix = np.zeros((n_val, n_models))
 
         for j, name in enumerate(self.models):
-            scores = self._compute_scores(y, preds_dict[name])
+            if self.metric is not None:
+                scores = self._compute_scores(y, preds_dict[name])
+            else:
+                scores = np.asarray(preds_dict[name])
+                # Validate shape
+                if scores.ndim != 1 or len(scores) != n_val:
+                    raise ValueError(
+                        f"Raw scores for model '{name}' must be a 1D array of length {n_val}; "
+                        f"got shape {scores.shape}."
+                    )
             self.matrix[:, j] = scores if self.mode == 'max' else -scores
 
         if self.task == 'classification':

{deskit-1.3 → deskit-1.3.2}/src/deskit/des/lwsei.py

@@ -3,6 +3,7 @@ LWSE-I: Locally Weighted Stacking Ensemble (Inverse-distance).
 """
 from deskit.base.predictbase import PredictBase
 from deskit._config import make_finder
+from deskit.base.knnbase import _drop_self_match
 from scipy.optimize import nnls
 import numpy as np
 
@@ -77,6 +78,10 @@ class LWSEI(PredictBase):
         self._y_val = y
         self._finder.fit(features)
 
+        # Required for PredictBase.predict() when hard labels are used
+        if self.task == 'classification':
+            self.classes_ = np.unique(y)
+
     def _weights_batch(self, x, temperature=None, k=None, loo=False, **kwargs):
         """
         Core weight computation. x is a 2-D float64 numpy array (batch, n_features).
@@ -127,4 +132,33 @@ class LWSEI(PredictBase):
             total = coeffs.sum()
             weights_out[b] = coeffs / total if total > 1e-10 else uniform
 
-        return weights_out
+        return weights_out
+
+    def _kneighbors(self, x, k=None, loo=False):
+        """
+        Query the fitted neighbor index, with optional leave-one-out (LOO)
+        exclusion of each query point's own occurrence in the DSEL.
+
+        Parameters
+        ----------
+        x : np.ndarray, shape (batch, n_features)
+        k : int, optional
+            Neighborhood size. None defers to the finder's default.
+        loo : bool
+            If True, query one extra neighbor per row and drop the
+            zero-distance match (the point itself) when present.
+
+        Returns
+        -------
+        distances, indices : np.ndarray, each shape (batch, k_eff)
+        """
+        if not loo:
+            return self._finder.kneighbors(x, k=k)
+
+        # Resolve default k if not given
+        if k is None:
+            probe_distances, _ = self._finder.kneighbors(x, k=k)
+            k = probe_distances.shape[1]
+
+        distances, indices = self._finder.kneighbors(x, k=k + 1)
+        return _drop_self_match(distances, indices, k)

{deskit-1.3 → deskit-1.3.2}/src/deskit/des/lwseu.py

@@ -3,6 +3,7 @@ LWSE-U: Locally Weighted Stacking Ensemble (Uniform).
 """
 from deskit.base.predictbase import PredictBase
 from deskit._config import make_finder
+from deskit.base.knnbase import _drop_self_match
 from scipy.optimize import nnls
 import numpy as np
 
@@ -77,6 +78,10 @@ class LWSEU(PredictBase):
         self._y_val = y
         self._finder.fit(features)
 
+        # Required for PredictBase.predict() when hard labels are used
+        if self.task == 'classification':
+            self.classes_ = np.unique(y)
+
     def _weights_batch(self, x, temperature=None, k=None, loo=False, **kwargs):
         """
         Core weight computation. x is a 2-D float64 numpy array (batch, n_features).
@@ -119,4 +124,33 @@ class LWSEU(PredictBase):
             total = coeffs.sum()
             weights_out[b] = coeffs / total if total > 1e-10 else uniform
 
-        return weights_out
+        return weights_out
+
+    def _kneighbors(self, x, k=None, loo=False):
+        """
+        Query the fitted neighbor index, with optional leave-one-out (LOO)
+        exclusion of each query point's own occurrence in the DSEL.
+
+        Parameters
+        ----------
+        x : np.ndarray, shape (batch, n_features)
+        k : int, optional
+            Neighborhood size. None defers to the finder's default.
+        loo : bool
+            If True, query one extra neighbor per row and drop the
+            zero-distance match (the point itself) when present.
+
+        Returns
+        -------
+        distances, indices : np.ndarray, each shape (batch, k_eff)
+        """
+        if not loo:
+            return self._finder.kneighbors(x, k=k)
+
+        # Resolve default k if not given
+        if k is None:
+            probe_distances, _ = self._finder.kneighbors(x, k=k)
+            k = probe_distances.shape[1]
+
+        distances, indices = self._finder.kneighbors(x, k=k + 1)
+        return _drop_self_match(distances, indices, k)

{deskit-1.3 → deskit-1.3.2/src/deskit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deskit
-Version: 1.3
+Version: 1.3.2
 Summary: A Python library for Dynamic Ensemble Selection
 Author: Tikhon Vodyanov
 License-Expression: MIT
@@ -251,6 +251,8 @@ router = DEWSU(task="regression", metric=pinball, mode="min", k=20)
 
 Built-in metric strings: `accuracy`, `mae`, `mse`, `rmse`, `log_loss`, `prob_correct`.
 
+Alternatively, pass `metric=None` and populate the dictionary passed to fit() with your own per-sample computed scores instead of predictions.
+
 ---
 
 ## Data types