added SparseCategoricalCrossEntropy Loss Function

MLlib/loss_func.py

@@ -1,5 +1,6 @@
 import numpy as np
 from MLlib.activations import sigmoid
+from MLlib.activations import softmax
 
 
 class MeanSquaredError():
@@ -34,6 +35,28 @@ def loss(X, Y, W):
         M = X.shape[0]
         return np.sum(np.absolute(np.dot(X, W).T - Y)) / M
 
+class SparseCategoricalCrossEntropy():
+    @staticmethod
+    def loss(X,Y,W,n):
+        # n = total number of classes for classification
+        # W of dimension (X.shape[1],n)
+        Yprime = []
+        for ydash in Y:
+            for y in ydash:
+                y=int(y)
+                yprime = list([0]*y+[1]+[0]*(n-y-1))
+                Yprime.append(yprime)
+        Yprime = np.array(Yprime)
+        H=np.dot(X,W)
+        for i in range(len(H)):
+            H[i]=softmax(H[i])
+        loss=0
+        for y in range(len(Yprime)):
+            for feature_index in range(len(Yprime[y])):
+                if Yprime[y][feature_index] ==1:
+                    loss-=np.log(H[y][feature_index])
+        return loss
+
     @staticmethod
     def derivative(X, Y, W):
         M = X.shape[0]