jinwoongkim / cs231n Goto Github PK

for i in xrange(num_train):
   scores = X[i].dot(W) # scores to classes e.g., [ 3.2, 1.3, 2.2] for cat, car, frog
   score_yi = scores[y[i]] # score for input e.g., 3.2 if X[i] is cat
   for j in xrange(num_classes):
     if j == y[i]:
       continue
     loss += max(0,scores[j] - score_yi + 1) 

 dW = np.zeros(W.shape) # initialize the gradient as zero
 # compute the loss and the gradient
 num_classes = W.shape[1]
 num_train = X.shape[0]
 loss = 0.0
 for i in xrange(num_train):
   scores = X[i].dot(W)
   score_yi = scores[y[i]]
   number_of_loss_contributor = 0
   for j in xrange(num_classes):
     if j == y[i]:
       continue
     margin = scores[j] - score_yi + 1 # note delta = 1
     if margin > 0:
       loss += margin
       dW[:, j] += X[i, :]
       number_of_loss_contributor += 1
   dW[:, y[i]] += (number_of_loss_contributor * -X[i])

 # Right now the loss is a sum over all training examples, but we want it
 # to be an average instead so we divide by num_train.
 loss /= num_train
 dW /= num_train

 # Add regularization to the loss.
 loss += reg * np.sum(W * W)
 dW += 2 * reg * W

 return loss, dW

  loss = 0.0
  for i in xrange(num_train):
    scores = X[i].dot(W)
    score_yi = scores[y[i]]
    sum_esj = 0
    for j in xrange(num_classes):
      if j == y[i]:
        continue
      sum_esj += np.exp(scores[j])
    loss += np.log(np.exp(score_yi)/sum_esj)*-1

    ### Compute the forward pass
    H1 = X@W1+b1
    H2 = H1@W2+b2
    print(H2)

    ### Compute the forward pass
    H1 = np.maximum(0,X@W1+b1)
    H2 = H1@W2+b2
    print(H2)