deskit 1.2.6__tar.gz → 1.2.8__tar.gz

{deskit-1.2.6/src/deskit.egg-info → deskit-1.2.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deskit
-Version: 1.2.6
+Version: 1.2.8
 Summary: A Python library for Dynamic Ensemble Selection
 Author: Tikhon Vodyanov
 License-Expression: MIT

{deskit-1.2.6 → deskit-1.2.8}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "deskit"
-version = "1.2.6"
+version = "1.2.8"
 description = "A Python library for Dynamic Ensemble Selection"
 readme = "README.md"
 license = "MIT"

{deskit-1.2.6 → deskit-1.2.8}/src/deskit/base/knnbase.py

@@ -57,6 +57,11 @@ class KNNBase(PredictBase, BaseRouter):
             scores = self._compute_scores(y, preds_dict[name])
             self.matrix[:, j] = scores if self.mode == 'max' else -scores
 
+        if self.task == 'classification':
+            self.classes_ = np.unique(y)
+        else:
+            self.classes_ = None
+
         self.model.fit(features)
 
     def _kneighbors(self, x, k=None, loo=False):

{deskit-1.2.6 → deskit-1.2.8}/src/deskit/base/predictbase.py

@@ -95,20 +95,25 @@ class PredictBase:
             Batch → shape ``(n_samples,)`` or ``(n_samples, n_classes)``.
         """
         x, batch_size = _prepare(X_test)
-        weights = self._weights_batch(x, temperature=temperature, **kwargs)  # (batch, n_models)
+        weights = self._weights_batch(x, temperature=temperature, **kwargs)
 
         preds_list = [np.asarray(test_preds[m], dtype=float) for m in self.models]
         first = preds_list[0]
 
         if self.task == 'classification':
-            # Probability arrays: blend per-class columns.
-            # preds_3d : (batch, n_models, n_classes)
-            preds_3d = np.stack(preds_list, axis=1)
-            result = np.einsum("bm,bmc->bc", weights, preds_3d)  # (batch, n_classes)
+            if first.ndim == 1:
+                # Hard labels: convert to one-hot internally
+                n_classes = len(self.classes_)
+                preds_2d = np.stack(preds_list, axis=1)                 # (batch, n_models)
+                preds_3d = np.eye(n_classes)[preds_2d.astype(int)]      # (batch, n_models, n_classes)
+            else:
+                preds_3d = np.stack(preds_list, axis=1)                 # (batch, n_models, n_classes)
+            result = np.einsum("bm,bmc->bc", weights, preds_3d)         # (batch, n_classes)
+            if first.ndim == 1:
+                result = np.argmax(result, axis=1)                      # (batch,)
         else:
-            # Scalar predictions: weighted average.
-            preds_2d = np.stack(preds_list, axis=1)              # (batch, n_models)
-            result = (weights * preds_2d).sum(axis=1)             # (batch,)
+            preds_2d = np.stack(preds_list, axis=1)
+            result = (weights * preds_2d).sum(axis=1)
 
         return result[0] if batch_size == 1 else result
 

{deskit-1.2.6 → deskit-1.2.8/src/deskit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deskit
-Version: 1.2.6
+Version: 1.2.8
 Summary: A Python library for Dynamic Ensemble Selection
 Author: Tikhon Vodyanov
 License-Expression: MIT