Clash Royale CLAN TAG#URR8PPP

How to gather the logits according to a 2D index?

Assume the sequence [[2, 1, 4], [3, 4, 2]] is generated by a pre-trained LSTM. It's dimension is (2*3) meaning batch-size = 2 and 3 time steps in each sample.

[[2, 1, 4], [3, 4, 2]]

(2*3)

batch-size = 2

3 time steps

Then for example, there are 5 features in total so the logits may be:

5 features

[[[0.2, 0.2, 0.1, 0.1, 0.2],
[0.3, 0.2, 0.1, 0.1, 0.1],
[0.1, 0.2, 0.1, 0.1, 0.3]],
[[0.2, 0.2, 0.1, 0.1, 0.2],
[0.2, 0.2, 0.1, 0.1, 0.2],
[0.2, 0.2, 0.1, 0.1, 0.2]]]


I want to use the sequence as the index to get the corresponding probabilities from the logits for each sample and each time step. Regarding the example above, the final result I want to get is

[[0.1, 0.2, 0.3],[0.1, 0.2, 0.1]]


I knew that I probably need tf.stack() but I'm confused about how to handle the dimension. Appreciate for any help!

1 Answer
1

I think I found a way.

tf.losses.sparse_softmax_cross_entropy(labels = None, logits = None)


labels is the index and logits is the output of a model.

labels

logits

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.