Tensorflow Log Loss功能最终成为一个向量？

Question

我正在尝试在tensorflow中实现多类日志丢失功能。以下是我提出的建议：

    #self.sampled_actions:     Tensor [batchsize, 2]   [[0,1], [1,0], .......] 
    #                          which is one hot encoded
    #self.probability:         Tensor [batchsize, 2]   [[.4,.6], [.3,.7], .....]
    #                          Represents probability of each class by neural net output
    #self.discounted_rewards   Tensor [batchsize, 1]   [0.4 0.5 0.5 -0.1, .....]
    #                          Differenc weights for different data points

    self.batch_loss = tf.log(tf.reduce_sum(self.sampled_actions * self.probability, axis=1))*self.discounted_rewards
    self.loss = -tf.reduce_sum(  self.batch_loss , axis=0)

然而，我的损失函数最终是一个向量而不是单个值。我出错的地方？

Answer 1

删除轴= 0。 请从正式的TensorFlow站点查看reduce_sum的一些示例：

x = tf.constant([[1, 1, 1], [1, 1, 1]])
tf.reduce_sum(x)  # 6
tf.reduce_sum(x, 0)  # [2, 2, 2]
tf.reduce_sum(x, 1)  # [3, 3]
tf.reduce_sum(x, 1, keepdims=True)  # [[3], [3]]
tf.reduce_sum(x, [0, 1])  # 6