我正在学习神经网络,并且我想在Python中编写一个cross_entropy函数。它的定义如下:
其中N是样本数量,k是类别数量,log是自然对数,t_i,j在样本i属于类别j时为1,否则为0,p_i,j是样本i属于类别j的预测概率。为了避免对数计算中的数值问题,将预测值限制在[10^{−12}, 1 − 10^{−12}]范围内。
根据上述描述,我编写了代码,将预测值限制在[epsilon, 1 − epsilon]范围内,然后根据上述公式计算交叉熵:
def cross_entropy(predictions, targets, epsilon=1e-12):    """    Computes cross entropy between targets (encoded as one-hot vectors)    and predictions.     Input: predictions (N, k) ndarray           targets (N, k) ndarray            Returns: scalar    """    predictions = np.clip(predictions, epsilon, 1. - epsilon)    ce = - np.mean(np.log(predictions) * targets)     return ce以下代码将用于检查cross_entropy函数是否正确:
predictions = np.array([[0.25,0.25,0.25,0.25],                        [0.01,0.01,0.01,0.96]])targets = np.array([[0,0,0,1],                  [0,0,0,1]])ans = 0.71355817782  #正确答案x = cross_entropy(predictions, targets)print(np.isclose(x,ans))上述代码的输出为False,也就是说我定义cross_entropy函数的代码不正确。然后我打印了cross_entropy(predictions, targets)的结果。它给出了0.178389544455,而正确的结果应该是ans = 0.71355817782。谁能帮我检查一下我的代码有什么问题?
回答:
你离正确答案并不远,但请记住,你是在计算N个和的平均值,其中N=2(在这种情况下)。所以你的代码可以改成:
def cross_entropy(predictions, targets, epsilon=1e-12):    """    Computes cross entropy between targets (encoded as one-hot vectors)    and predictions.     Input: predictions (N, k) ndarray           targets (N, k) ndarray            Returns: scalar    """    predictions = np.clip(predictions, epsilon, 1. - epsilon)    N = predictions.shape[0]    ce = -np.sum(targets*np.log(predictions+1e-9))/N    return cepredictions = np.array([[0.25,0.25,0.25,0.25],                        [0.01,0.01,0.01,0.96]])targets = np.array([[0,0,0,1],                   [0,0,0,1]])ans = 0.71355817782  #正确答案x = cross_entropy(predictions, targets)print(np.isclose(x,ans))在这里,我认为使用np.sum()会更清晰一些。另外,我在np.log()中添加了1e-9,以避免计算中出现log(0)的可能。希望这对你有帮助!
注意:正如@Peter的评论所指出的,如果你的epsilon值大于0,那么1e-9的偏移确实是多余的。
