what to do when `ppl=-nan`?

[geovedi]

so I experimenting subwords units to minimise vocabs, but always end-up with ppl=-nan.

tried to avoid it is by making sure the unbalanced data removed using tools like filter-length.pl and lowering learning rate, but still no luck.

[neubig]

If you can provide the data and exact command that you ran, I might be able to debug. Graham On Jan 6, 2017 4:32 AM, "Jim Geovedi" <notifications@github.com> wrote: so I experimenting subwords units to minimise vocabs, but always end-up with ppl=-nan. tried to avoid it is by making sure the unbalanced data removed using tools like filter-length.pl and lowering learning rate, but still no luck. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAYWG-aHLPTG_jSJbUmXsxs8P-5m6eO-ks5rPgoXgaJpZM4Lci9s> .

[geovedi]

sure, i'll send it privately to your @cs.cmu.edu e-mail address.

[geovedi]

looks like i have a bad data. works fine on my other data.