why ksize of maxpooling of textcnn is [1, self.sentence_size - filter_size + 1, 1, 1],what I mind is "sentence_size - filter_size + 1", can it be sentence_size, filter_size or any others, is there any tricks? I couldn't see it in paper.
Since the convolutional layer of CNN use the "VALID" padding, the dimension of the convolutional output is "sentence_size - filter_size + 1". That's why use "sentence_size - filter_size + 1".
why ksize of maxpooling of textcnn is [1, self.sentence_size - filter_size + 1, 1, 1],what I mind is "sentence_size - filter_size + 1", can it be sentence_size, filter_size or any others, is there any tricks? I couldn't see it in paper.