0

I am using an AutoML Natural Language model to classify articles into different categories on Google cloud. Is there a way to continuously train this model with new data so that it can improve itself?

Or do I have to train a new model after some time with new data manually and then replace the current model with it?

Shubham Sharma
  • 1
  • Have you read about [Evaluation](https://cloud.google.com/natural-language/automl/docs/evaluate) concept? Indicating the accuracy of the trained model you can every time make decision what can be improved. – Nick_Kh Sep 04 '20 at 09:19
  • @Nick_Kh I am talking about Google Cloud's AutoML models. It is not possible to hyper tune these models manually. my question is regarding continuous training - by continuous training I mean the model automatically improves itself by training on new data. – Shubham Sharma Sep 05 '20 at 10:16
  • Every time you aimed to train the model you need to supply dataset with some fresh data for further analysis, thus I would not call it continuous training, unless you're seeking how to automate this process. Have you seen this [thread](https://stackoverflow.com/questions/57910005/automating-the-google-cloud-automl-pipeline)? – Nick_Kh Sep 07 '20 at 12:29

1 Answers1

0

I was looking for this option too, unfortunately Google doesn't provide such an option. All I could find is the option of retraining the model as a new version of the existing one (with new or added data), and compare their results. retraining as a new version - pic

Chris Schaller
  • 13,704
  • 3
  • 43
  • 81
Michal Ashkenazi
  • 1