Use the last char instead of the first in prediction

[leondgarse]

1. As it's mentioned the start_string could be a string here, in my understanding, the correctly predicted next character should be the last one in predictions.
2. As tf.multinomial is not used any more here, temperature has no effect either.

• i.e.
  • With start_string = 'QUEEN', the model prediction string is UEEN:, so the next char should be :
  • In the loop run for i in range(num_generate):
    • With predicted_id = tf.argmax(predictions[0]).numpy(), output is QUEENUS:\nI think so ...
    • With predicted_id = tf.argmax(predictions[-1]).numpy(), output is QUEEN:\nThe grand the ...

[alextp]

@yashk2810 does this make sense to you?

I'm a little confused about what this change entails.

[yashk2810]

So, I tried out the change and for a single character start_string, the predictions shape is (1, 65). But if the start_string is a word like 'Queen', the shape of the first prediction is (5, 65). Hence we need to use the last character from the prediction to predict the next character, hence predictions[-1] makes sense there.

This doesn't matter if the input is a single character. So changing to predictions[-1] will make the predictions right if the start_string is a word and it won't change anything if the start_string is a character.

As far as temperature goes, it makes sense to remove it since tf.multinomial is not being used.

@leondgarse Can you also remove the comment above the predicted_id line which mentions tf.multinomial and change the text in the markdown above?

[leondgarse]

Sorry for the delay, I just replaced the sentence a multinomial distribution by argmax in comment and the above text.

[tensorflowbutler]

Nagging Assignee @yifeif: It has been 14 days with no activity and this issue has an assignee. Please update the label and/or status accordingly.