cancel
Showing results for 
Search instead for 
Did you mean: 

refreshing models on schedule?

Highlighted
Image Sensor

Hello. How often should I expect to have to retrain/refresh my models? Is that something I can anticipate and schedule for?

Labels (3)
1 Reply
Highlighted
Blue LED

Hey @annapeters0n ,

The short answer is it depends on your model usage. Here is a quick article to read on model maintenance: https://towardsdatascience.com/why-your-models-need-maintenance-faff545b38a2

Luckily, Data Robot has a solution to help figure out when you should probably re-train your model through the "Data Drift" feature. https://app.datarobot.com/docs/predictions/deploy/data-drift.html 

Data Robot can tell you which of your model deployments are "At Risk" or "Failing" for Data Drift and that is usually a good indicator that the model might need to be re-trained/refreshed/investigated. In our experience it is kind of hard to anticipate/schedule for in advance but you can setup alerts to notify you when the model is at risk/failing for data drift. One of the things our team has planned to work on is automating the process of getting a data drift alert from data robot and automatically re-train/deploy the model based on some additional criteria.