Log perplexity during LDA model fitting.

[groceryheist]

This is a feature request. I would like to find out if my model fit is converging on a hold out set. Currently eval_every operates on the latest mini-batch. A parameter ( perhaps called eval_corpus ) should be available to evaluate convergence on a hold-out set.

[parulsethi]

This is being worked on in #1399. It would be merged within this week but if urgent, you can clone my gensim fork here and checkout to tensorboard_logs branch for this functionality. This is a tutorial notebook with instructions to use the API. As it's a new feature, let us know if you have any additional suggestions too.

[groceryheist]

Thank you! Happy to see this being worked on. Not urgent for me. I have hacked a temporary solution for me and will update when you are done.

…

On Wed, Aug 23, 2017 at 7:20 PM, Parul Sethi ***@***.***> wrote: This is being worked on in #1399 <#1399>. It would be merged within this week but if urgent, you can clone my gensim fork here <https://github.com/parulsethi/gensim> and checkout to tensorboard_logs branch for this functionality. This <https://github.com/parulsethi/gensim/blob/tensorboard_logs/docs/notebooks/Training_visualizations.ipynb> is a tutorial notebook with instructions to use the API. As it's a new feature, let us know if you have any additional suggestions too. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1546 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAG6n5LM9NrbAeGCs--z_uS8eYm5i7W7ks5sbN3WgaJpZM4O_ie_> .

-- Nate

[menshikh-iv]

Implemented in #1399