CS231n-assignment1作业knn实现总结

knn实现部分重要的是距离的求法。作业中提到了三种求距离的方法。分别用两个循环，一个循环和无循环（用矩阵点积）。

两个循环的方法：

def compute_distances_two_loops(self, X):
    num_test = X.shape[0]  #500
    num_train = self.X_train.shape[0]  #5000
    dists = np.zeros((num_test, num_train))
    for i in range(num_test):
        for j in range(num_train):       
            dist=np.sqrt(np.sum(np.square(X[i]-self.X_train[j])))
            dists[i][j]=dist       
    return dists

单个循环的方法：

  def compute_distances_one_loop(self, X):
    num_test = X.shape[0]
    num_train = self.X_train.shape[0]
    dists = np.zeros((num_test, num_train))
    for i in range(num_test): 
        dist=np.sqrt(np.sum(np.square(X[i]-self.X_train),axis=1))
        dists[i,:]=dist     
    return dists

无循环的方法：

def compute_distances_no_loops(self, X):
    num_test = X.shape[0]
    num_train = self.X_train.shape[0]
    dists = np.zeros((num_test, num_train)) 
   
    # (x1 - y1)^2 + (x2 - y2)^2 = (x1^2 + x2^2) + (y1^2 + y2^2) - 2*(x1*y1 + x2*y2)
    train_sq = np.sum(self.X_train ** 2, axis=1, keepdims=True)  
    # (m, 1), 注意 keepdims 的含义
    train_sq = np.broadcast_to(train_sq, shape=(num_train, num_test)).T  # (n, m), 注意转置
    test_sq = np.sum(X ** 2, axis=1, keepdims=True)  # (n, 1)
    test_sq = np.broadcast_to(test_sq, shape=(num_test, num_train))  # (n, m)
    cross = np.dot(X, self.X_train.T)  # (n, m)
    dists = np.sqrt(train_sq + test_sq - 2 * cross)  # 开根号
    return dists

无循环求距离的推导过程：