SVM/hinge loss function

SVM/hinge loss function

loss function

CS231n课程作业一中，涉及到了SVM损失函数，经过研究，应该指的是hinge loss。其公式为：

$$L_i=\sum_{j\neq y_i}max(0,w_j^Tx_i-w_{y_i}^Tx_i+\Delta)$$

• 循环方式实现：

def svm_loss_naive(W, X, y, reg):
  """
  Structured SVM loss function, naive implementation (with loops).

  Inputs have dimension D, there are C classes, and we operate on minibatches
  of N examples.

  Inputs:
  - W: A numpy array of shape (D, C) containing weights.
  - X: A numpy array of shape (N, D) containing a minibatch of data.
  - y: A numpy array of shape (N,) containing training labels; y[i] = c means
    that X[i] has label c, where 0 <= c < C.
  - reg: (float) regularization strength

  Returns a tuple of:
  - loss as single float
  - gradient with respect to weights W; an array of same shape as W
  """
  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)
    correct_class_score = scores[y[i]]
    for j in xrange(num_classes):
      margin = scores[j] - correct_class_score + 1 # note delta = 1
      if margin > 0:
        if j != y[i]:
          loss += margin


  # 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

  # Add regularization to the loss.
  loss += 0.5 * reg * np.sum(W * W)

  return loss, dW

• 对应的向量化代码（自己实现）：

def svm_loss_vectorized(W, X, y, reg):
  """
  Structured SVM loss function, vectorized implementation.

  Inputs and outputs are the same as svm_loss_naive.
  """
  loss = 0.0
  dW = np.zeros(W.shape) # initialize the gradient as zero

  scores = X.dot(W)
  num_train = X.shape[0]
  rows = range(num_train)
  correct_class_score = scores[rows,y]
  margins = np.maximum(0,scores-np.reshape(correct_class_score,[num_train,1])+1)
  margins[rows,y] = 0
  loss = np.sum(margins)
  loss /= num_train
  loss += 0.5 * reg * np.sum(W * W)
  pass

  return loss, dW

gradient of loss function

对higne函数求导推导如下：
由公式$L_i=\sum_{j\neq y_i}max(0,w_j^Tx_i-w_{y_i}^Tx_i+\Delta)$，考虑两种情况，

• $j\neq y_i$:
  此时，对$w_j$求偏导，式$w_j^Tx_i-w_{y_i}^Tx_i+\Delta$中只有第一项有效果，第二项为常数。如果$w_j^Tx_i-w_{y_i}^Tx_i+\Delta>0$,那么求得的值为$x_i$；如果$w_j^Tx_i-w_{y_i}^Tx_i+\Delta\le0$，由于$max$的作用，函数返回常数0，故偏导数为0。综上，可知，此时的偏导为：
  $$\nabla_{w_{y_i}}L_i=1(w_j^Tx_i-w_{y_i}^Tx_i+\Delta>0)x_i$$

• $j=y_i$：
  此时，对$w_j$求偏导，式$w_j^Tx_i-w_{y_i}^Tx_i+\Delta$中只有第二项有效果，第一项为常数($j=y_i$)。如果$w_j^Tx_i-w_{y_i}^Tx_i+\Delta>0$,那么求得的值为$-x_i$；如果$w_j^Tx_i-w_{y_i}^Tx_i+\Delta\le0$，由于$max$的作用，函数返回常数0，故偏导数同样为0。需要注意的是，此时由于$\sum$的存在，导致偏导结果如下：
  

  $$\nabla_{w_{y_i}}L_i=-\left(\sum_{j\neq y_i}1(w_j^Tx_i-w_{y_i}^Tx_i+\Delta>0)\right)x_i$$

• 循环方式实现

def svm_loss_naive(W, X, y, reg):
  """
  Structured SVM loss function, naive implementation (with loops).

  Inputs have dimension D, there are C classes, and we operate on minibatches
  of N examples.

  Inputs:
  - W: A numpy array of shape (D, C) containing weights.
  - X: A numpy array of shape (N, D) containing a minibatch of data.
  - y: A numpy array of shape (N,) containing training labels; y[i] = c means
    that X[i] has label c, where 0 <= c < C.
  - reg: (float) regularization strength

  Returns a tuple of:
  - loss as single float
  - gradient with respect to weights W; an array of same shape as W
  """
  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)
    correct_class_score = scores[y[i]]
    for j in xrange(num_classes):
      margin = scores[j] - correct_class_score + 1 # note delta = 1
      if margin > 0:
        if j != y[i]:
          loss += margin
          dW[:, y[i]] += -1 * X[i]
          dW[:, j] += 1 * 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 += 0.5 * reg * np.sum(W * W)
  dW += reg*W

  return loss, dW

• 向量化实现

def svm_loss_vectorized(W, X, y, reg):
  """
  Structured SVM loss function, vectorized implementation.

  Inputs and outputs are the same as svm_loss_naive.
  """
  loss = 0.0
  dW = np.zeros(W.shape) # initialize the gradient as zero

  scores = np.dot(X,W)
  num_train = X.shape[0]
  rows = range(num_train)
  correct_class_score = scores[rows,y]
  margins = np.maximum(0,scores-np.reshape(correct_class_score,[num_train,1])+1)
  margins[rows,y] = 0
  loss = np.sum(margins)
  loss /= num_train
  loss += 0.5 * reg * np.sum(W * W)

  margins01 = 1 * (margins > 0)
  margins01[rows,y] = -1*np.sum(margins01, axis=1)
  dW = np.dot(X.transpose(), margins01)
  dW /= num_train
  dW += reg * W

  return loss, dW

---