Update loss_functions.py

code/loss_functions.py

@@ -1,35 +1,57 @@
 import math
 import numpy as np
 
-# using the common convention:
-#	yHat is the prediction while y is the truth (true label)
+### Note ###
+
+# yHat is prediction
+# y is the target (true label)
+
+
+### Functions ###
 
 def CrossEntropy(yHat, y):
-    pass
+  if yHat == 1:
+    return -log(y)
+  else:
+    return -log(1 - y)
+
+
+def Dice(yHat, y):
+  total = np.sum(y, dim=1) + np.sum(yHat, dim=1)
+  intersection = np.sum(y * yHat, dim=1)
+  dice = (2.0 * intersection) / (total + 1e-7)
+  return np.mean(dice)
 
 
 def Hinge(yHat, y):
-    pass
+  return np.max(0, 1 - yHat * y)
+
+
+def Huber(yHat, y):
+  pass
 
 
 def KLDivergence(yHat, y):
-    pass
+  pass
 
 
-def L1(yHat, y):s
-    return np.sum(np.absolute(yHat - y))
+def L1(yHat, y):
+  return np.sum(np.absolute(yHat - y))
 
 
 def L2(yHat, y):
-    return np.sum((yHat - y)**2)
+  return np.sum((yHat - y)**2)
 
 
 def MLE(yHat, y):
-    pass
+  pass
 
 
 def MSE(yHat, y):
-    return np.sum((yHat - y)**2) / y.size
+  return np.sum((yHat - y)**2) / y.size
+
+
+### Derivatives ###
 
 def MSE_prime(yHat, y):
     return yHat - y