Is training additive?


(Etl) #1

If I create a YML file or JSON file and load it with the /train HTTP API, and then I made a different training file and load it with the /train HTTP API. Are both training part of the last model, or it’s a new model each time, using only the training uploaded in that HTTP call?


(Souvik Ghosh) #2

It is a new model each time, what you are suggesting is a bit like transfer learning which usually don’t work for text the same way as it does for same for images


(Etl) #3

Thank you for clarifying. I mostly wanted to make sure I had understood this right.

So every time I want to train, I have to post all the training I created previously, not just the new ones.

That make sense. Thanks.