Training goes on infinitely

[naveen-kinnal]

Hello,

I see that the training for the YELP dataset goes on infinitely since the training in main.py has the condition 'while true'. May I know how many epochs should we train?

[jxhe]

Hi,

The training will end automatically when the val loss stops decreasing several times, as controlled here:

deep-latent-sequence-model/src/main.py

Line 743 in 8a79858

if args.patience >= 0:

You can reduce the patience number to make it end more quickly

[naveen-kinnal]

Thank you. I have another question

1. Why is the 'dev_src_file' and 'dev_trg_ref' the same?
2. If the src_file contains the negative sentences, the trg_file will contain the negative attributes?

[jxhe]

Can you clarify where "dev_src_file" and "dev_trg_ref" are?