2
$\begingroup$

I have trained a model for predicting the sale of items daily, such as daily car sales, with a machine learning model in Python. Now I get the new time real data (time series data). I want to retrain my mode. There are some issues:

  1. How much data I should use for retraining model, namely what's the time range of data or when I should retrain my model. For example use the new one day or one week et al., data to retrain model.

  2. What's condition of new data if I want to use the new data for retraining model.

  3. How to evaluate the stability of new training model, and How to know abnormal prediction for new model.

$\endgroup$

migrated from stackoverflow.com Aug 15 '17 at 11:29

This question came from our site for professional and enthusiast programmers.

1
$\begingroup$

Once you trained your model, you have a measure of accuracy.

You should retrain your model when the accuracy drops below a certain threshold.

Another approach is to use a Bayesian model, in which you update your model with each new observation.

I would suggest looking into pymc3's examples

$\endgroup$
  • $\begingroup$ thanks for your valuable comments. I will check the documents you mentioned. Could you tell me how to know abnormal prediction if I used the new retrained model. $\endgroup$ – tktktk0711 Aug 16 '17 at 1:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.