CS231n: Comment calculer le gradient de Softmax perte de fonction?

Je suis de regarder quelques vidéos pour Stanford CS231: Convolutifs Réseaux de Neurones pour la Reconnaissance Visuelle, mais ne comprends pas très bien comment calculer le gradient analytique pour softmax la perte de la fonction en utilisant numpy.

De cette stackexchange réponse, softmax gradient est calculé comme suit:

Python de mise en œuvre ci-dessus est:

num_classes = W.shape[0]
num_train = X.shape[1]
for i in range(num_train):
  for j in range(num_classes):
    p = np.exp(f_i[j])/sum_i
    dW[j, :] += (p-(j == y[i])) * X[:, i]

Quelqu'un pourrait-il expliquer comment l'extrait ci-dessus travail? De mise en œuvre détaillé pour softmax est également inclus ci-dessous.

def softmax_loss_naive(W, X, y, reg):
  """
  Softmax loss function, naive implementation (with loops)
  Inputs:
  - W: C x D array of weights
  - X: D x N array of data. Data are D-dimensional columns
  - y: 1-dimensional array of length N with labels 0...K-1, for K classes
  - reg: (float) regularization strength
  Returns:
  a tuple of:
  - loss as single float
  - gradient with respect to weights W, an array of same size as W
  """
  # Initialize the loss and gradient to zero.
  loss = 0.0
  dW = np.zeros_like(W)

  #############################################################################
  # Compute the softmax loss and its gradient using explicit loops.           #
  # Store the loss in loss and the gradient in dW. If you are not careful     #
  # here, it is easy to run into numeric instability. Don't forget the        #
  # regularization!                                                           #
  #############################################################################

  # Get shapes
  num_classes = W.shape[0]
  num_train = X.shape[1]

  for i in range(num_train):
    # Compute vector of scores
    f_i = W.dot(X[:, i]) # in R^{num_classes}

    # Normalization trick to avoid numerical instability, per http://cs231n.github.io/linear-classify/#softmax
    log_c = np.max(f_i)
    f_i -= log_c

    # Compute loss (and add to it, divided later)
    # L_i = - f(x_i)_{y_i} + log \sum_j e^{f(x_i)_j}
    sum_i = 0.0
    for f_i_j in f_i:
      sum_i += np.exp(f_i_j)
    loss += -f_i[y[i]] + np.log(sum_i)

    # Compute gradient
    # dw_j = 1/num_train * \sum_i[x_i * (p(y_i = j)-Ind{y_i = j} )]
    # Here we are computing the contribution to the inner sum for a given i.
    for j in range(num_classes):
      p = np.exp(f_i[j])/sum_i
      dW[j, :] += (p-(j == y[i])) * X[:, i]

  # Compute average
  loss /= num_train
  dW /= num_train

  # Regularization
  loss += 0.5 * reg * np.sum(W * W)
  dW += reg*W

  return loss, dW

OriginalL'auteur Nghia Tran | 2017-01-15