Brain.js add training to initial training without completely retraining

Question

I am currently using Brain.js in a project and at the moment it has to be trained all in one go,

Use train() to train the network with an array of training data. The network has to be trained with all the data in bulk in one call to train().

Which means if you want to add some new data later to improve it you have to retrain from the very beginning which is obviously not ideal.

Is there an option similar to brain.js which allows retraining without having to start from the beginning, or is anyone currently working on altering brain.js to allow this, or is this even possible (or would brain.js have to be completely rewritten), and if so would it be particularly difficult for someone new to javascript and programming in general (me) to add this in, and where would i even start.

Thanks for any help

Also, could someone with over 1500 reputation make a tag for brain.js, thanks

score 3 · Accepted Answer · answered Oct 15 '13 at 15:07

3

The documentation seems to imply that any new information requires a complete retrain. https://github.com/nickpoorman/brain Nick Poorman's fork seems to have the ability to stream train, which has what you are looking for, it has yet to be merged with the main repository but it has a pull request.

answered Oct 15 '13 at 15:07

VoronoiPotato

3,113
20
30

1

Thankyou, i think this is what i'm looking for but i'm not completely sure, does this mean that i can train the network, then use it, then continue training where it was stopped? – Ryan White Oct 16 '13 at 02:02
I used to know the answer to that question. I believe you are correct. Sorry I'm a bit late in responding. – VoronoiPotato Apr 30 '19 at 21:56
You cannot do it with brain.js and RNN LSTM it is not supported. – Haddar Macdasi May 08 '19 at 13:21

Brain.js add training to initial training without completely retraining

1 Answers1