keras model.fit函数需要哪种数据格式？

Question

I'm would like to know which data format the model.fit function of keras needs. The documentation is not specific enough for me. So it seems, that for an LSTM model it needs a 3D array for the parameter x. Some more specific questions: Does the data format depend on the chosen model? What is the meaning of each dimension of x? And what is the meaning of y?

Thanks in advance for anybody who can tell me a bit about that! Holger

Answer 1

The data format certainly depends on the model. You can have models that have multiple inputs, such as Siamese networks.

In the case of an LSTM, I believe the input is 2-D as in this example . That example loads data from the IMDB dataset . The relevant line of code there is:

xs = [[oov_char if (w >= num_words or w < skip_top) else w for w in x] for x in xs]

The first dimension corresponds to different examples, and the second dimension is the timestep.

As for y, that refers to the labels. In a sequence to sequence example, this would also be two dimensional with the same [example_index, timestep] indexing. However, in classification it be 1-dimensional with one label for each example.