Update _Dist/NeuralNetworks/c_BasicNN

Author: carefree0910

_Dist/NeuralNetworks/c_BasicNN/NNCore.py

@@ -633,5 +633,9 @@ def predict(self, x, get_raw=False, verbose=True):
         tensor = "_output" if self.is_regression or get_raw else "_prob_output"
         return self._calculate(x, tensor, "Predict", verbose=verbose)
 
+    def predict_classes(self, x, get_raw=False, verbose=True):
+        pred = self.predict(x, get_raw, verbose)
+        return np.argmax(pred, axis=1)
+
     def evaluate(self, x, y, verbose=False):
         print("{}: {:8.6}".format(self.metric_name, self.metric(y, self.predict(x, verbose=verbose))))

_Dist/NeuralNetworks/c_BasicNN/NNUtils.py

@@ -1,5 +1,7 @@
 import numpy as np
 import tensorflow as tf
+
+from scipy import interp
 from sklearn import metrics
 
 
@@ -28,6 +30,8 @@ class Metrics:
         "mse": -1, "ber": -1,
         "log_loss": -1
     }
+    require_prob = {name: False for name in sign_dict}
+    require_prob["auc"] = True
 
     @staticmethod
     def check_shape(y, binary=False):
@@ -54,6 +58,19 @@ def auc(y, pred):
             Metrics.check_shape(pred, True)
         )
 
+    @staticmethod
+    def multi_auc(y, pred):
+        n_classes = y.shape[1]
+        fpr, tpr = [None] * n_classes, [None] * n_classes
+        for i in range(n_classes):
+            fpr[i], tpr[i], _ = metrics.roc_curve(y[:, i], pred[:, i])
+        new_fpr = np.unique(np.concatenate([fpr[i] for i in range(n_classes)]))
+        new_tpr = np.zeros_like(new_fpr)
+        for i in range(n_classes):
+            new_tpr += interp(new_fpr, fpr[i], tpr[i])
+        new_tpr /= n_classes
+        return metrics.auc(new_fpr, new_tpr)
+
     @staticmethod
     def acc(y, pred):
         return np.mean(Metrics.check_shape(y) == Metrics.check_shape(pred))