I don't know which one is the better way to do. I use scikit to train ML and save it.

I train Model M with train data X and test data y, then save it to predict in next time.

When I get new observation data X_new, how should I retrain with X_new? (Assume the X_new is normal distribution but size is half of X)

I don't know is it ok to retrain only with X_new or need to retrain with X+X_new?

I know loading M and train X+X_new is the best way.

But if I lose train data X? That's what I concerned most.

Is it effective retraining only with X_new? the previous coef in M will be affected?


  • $\begingroup$ If X_new is roughly similar to X then you should get similar results, otherwise you could get very different results. It would be best to not lose that data. Everytime you retrain you will get different coefficients (depending on how different the new data is). $\endgroup$ – user2974951 Dec 5 '18 at 13:06
  • $\begingroup$ @user2974951 so the model can't memorize the trained state of last time and continuous training only with X_new, right? $\endgroup$ – code_worker Dec 6 '18 at 1:54
  • $\begingroup$ Most of the time, no. $\endgroup$ – user2974951 Dec 6 '18 at 6:27

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.