predict_proba() now sums to 1

dbgsom/BaseSom.py

@@ -93,7 +93,8 @@ def fit(self, X: npt.ArrayLike, y: None | npt.ArrayLike = None):
         """
         # Horizontal growing phase
 
-        X, y = self._prepare_inputs(X, y)
+        X, y = self._check_input_data(X, y)
+        # self._fit(X, y)
         if y is not None:
             classes, y = np.unique(y, return_inverse=True)
             self.classes_ = np.array(classes)
@@ -116,16 +117,11 @@ def fit(self, X: npt.ArrayLike, y: None | npt.ArrayLike = None):
         self.labels_ = self.predict(X)
         self.n_iter_ = self._current_epoch
 
-        self._fit(X, y)
-
         return self
 
-    def _prepare_inputs(self, X, y):
+    def _check_input_data(self, X, y):
         raise NotImplementedError
 
-    def _fit(self, X, y):
-        pass
-
     def predict(self, X):
         raise NotImplementedError
 
@@ -921,9 +917,8 @@ def linear_decay(
     learning_rate=None,
 ) -> float:
     """Linear decay between sigma_start and sigma_end over t training iterations."""
-    sigma = sigma_start * (1 - current_iter / max_iter) + sigma_end * (
-        current_iter / max_iter
-    )
+    ratio = current_iter / max_iter
+    sigma = sigma_start * (1 - ratio) + sigma_end * ratio
 
     return sigma
 

dbgsom/SomClassifier.py

@@ -12,7 +12,6 @@
     check_is_fitted,
     check_X_y,
 )
-
 from dbgsom.BaseSom import BaseSom
 
 
@@ -121,7 +120,7 @@ class SomClassifier(BaseSom, TransformerMixin, ClassifierMixin):
         Average distance from all training samples to their nearest prototypes.
     """
 
-    def _prepare_inputs(
+    def _check_input_data(
         self, X: npt.ArrayLike, y=npt.ArrayLike
     ) -> tuple[npt.NDArray, npt.ArrayLike]:
         X, y = check_X_y(X=X, y=y, ensure_min_samples=4, dtype=[np.float64, np.float32])
@@ -151,7 +150,8 @@ def _label_prototypes(self, X, y) -> None:
                 )
 
     def _fit(self, X: npt.ArrayLike, y: None | npt.ArrayLike = None):
-        self.classes_, y = np.unique(y, return_inverse=True)
+        classes, y = np.unique(y, return_inverse=True)
+        self.classes_ = classes
 
     def predict(self, X: npt.ArrayLike) -> np.ndarray:
         """Predict class labels for samples in X.
@@ -208,6 +208,10 @@ def predict_proba(self, X: npt.ArrayLike) -> np.ndarray:
         else:
             X_transformed = self.transform(X)
             node_probabilities = self._extract_values_from_graph("probabilities")
-            sample_probabilities = X_transformed @ node_probabilities
+            # Sample Probabilities do not sum to 1
+            sample_probabilities_unnormalized = X_transformed @ node_probabilities
+            sample_probabilities = sample_probabilities_unnormalized / (
+                sample_probabilities_unnormalized.sum(axis=1)[np.newaxis].T
+            )
 
         return sample_probabilities

dbgsom/SomVQ.py

@@ -113,7 +113,7 @@ class SomVQ(BaseSom, ClusterMixin, TransformerMixin):
         Average distance from all training samples to their nearest prototype.
     """
 
-    def _prepare_inputs(self, X: npt.ArrayLike, y=None) -> tuple[npt.NDArray, None]:
+    def _check_input_data(self, X: npt.ArrayLike, y=None) -> tuple[npt.NDArray, None]:
         X = check_array(array=X, ensure_min_samples=4, dtype=[np.float64, np.float32])
         return X, y