add fixes for multi-class

pyChemometrics/ChemometricsPLSDA.py

@@ -228,20 +228,28 @@ def fit(self, x, y, **fit_params):
 
             else:
                 y_pred = self.predict(x)
-                accuracy = metrics.accuracy_score(y, y_pred)
-                precision = metrics.precision_score(y, y_pred, average='weighted')
-                recall = metrics.recall_score(y, y_pred, average='weighted')
-                misclassified_samples = np.where(y.ravel() != y_pred.ravel())[0]
-                f1_score = metrics.f1_score(y, y_pred, average='weighted')
-                conf_matrix = metrics.confusion_matrix(y, y_pred)
-                zero_oneloss = metrics.zero_one_loss(y, y_pred)
+
+                # Make dummy matrix into a label vector for scoring
+                if isDummy:
+                    # y = np.where(y == 1)[1]
+                    y_vec = np.where(y == 1)[1]
+                else:
+                    y_vec = y
+
+                accuracy = metrics.accuracy_score(y_vec, y_pred)
+                precision = metrics.precision_score(y_vec, y_pred, average='weighted')
+                recall = metrics.recall_score(y_vec, y_pred, average='weighted')
+                misclassified_samples = np.where(y_vec.ravel() != y_pred.ravel())[0]
+                f1_score = metrics.f1_score(y_vec, y_pred, average='weighted')
+                conf_matrix = metrics.confusion_matrix(y_vec, y_pred)
+                zero_oneloss = metrics.zero_one_loss(y_vec, y_pred)
                 matthews_mcc = np.nan
                 roc_curve = list()
                 auc_area = list()
 
                 # Generate multiple ROC curves - one for each class the multiple class case
                 for predclass in range(self.n_classes):
-                    current_roc = metrics.roc_curve(y, class_score[:, predclass], pos_label=predclass)
+                    current_roc = metrics.roc_curve(y_vec, class_score[:, predclass], pos_label=predclass)
                     # Interpolate all ROC curves to a finite grid
                     #  Makes it easier to average and compare multiple models - with CV in mind
                     tpr = current_roc[1]