Vector encode input extend vocab about pointer_summarizer HOT 3 CLOSED

The processing for each example is independent. The additional OOV vocabulary is created for pointers only, so it should only account for that particular oov. Looking at the model code would make it more clear

from pointer_summarizer.

I know you process each example independently, but with any first article_oov_word in any example, you would assign its index by len(vocab) + 1 no matter what is its index in another example. For example:
You have 2 sentences in a batch:

1. Give me the direction to my house
2. Tell me about alexa

And your vocab: [give, me, the, direction, to, my, tell, about], len = 8
After processing, 2 extended_vocab would be:

1. [0, 1, 2, 3, 4, 5, 8] --> 8 is for house
2. [6, 1, 7, 8, -1, -1, -1] --> 8 is for alexa

Is it ok for training?

from pointer_summarizer.

Yeh that’s how the pointer network work OOV is added to vocabulary just to point in that particular example. The purpose is NOT to extend the overall vocabulary. It would be more clear from the paper.

from pointer_summarizer.