merge two models in machine learning

Question

I'm currently training a ner model using stanford ner. Is it possible to merge two models together? or is there a way to expend current model?

Suppose I originally have a large corpus called A, and I get a model MA. (Which works good) But the problem is, it takes long time to train it and in the future I want to add entries into corpus A, say, B. I have to retrain (A + B) again which takes longer time.

Is there a way to just train B and merge it with A? Or I can.. expend the model MA to make it work for new entities.

Thanks!

Answer 1

There is no "general way". It depends on the models you are using, some of them can be trained in online mode (where new instancs, new data do not require retraining) while some cannot. So on this level of generality the answer to your question is no, you cannot, you need to retrain it each time. On the other hand, some of the models can be trained this way - look for those which support online learning.