Reiterating ULMFIT language model

My problem statement is in such a way that,

I will get text documents in batches(100-1000 approx), so I will need to retrain the model every time when I get the data.

So in the case of ULMFIT, do I need to retrain the language model also in every iteration? If yes, How can I do that?

Can I do that like,

  • load previous model,

  • add present data to the model

  • then fit the language model ??

Please help.

My assumption is that I have to retrain both the language model and text learner in every iteration? Please correct me if I’m wrong.

have anyone tried that? i would be interested in this also

To me, it would depend on some criteria:

Are your new articles introducing new languages/phrases that previous articles didn’t have? Then re-train the initial language model too.

If not, just train the classifier.

1 Like

hey @muellerzr,

  1. No, languages will not come but phrases may come
  2. If I have to train just the classifier, my doubt is that would the model’s previous weights be persistent when I’m retraining it with new data? If yes, can you please guide me on how to achieve it if you have code sample or some links? Thank you so much

No @rubenarana, I couldn’t solve it.