In my project, i trained network with 80% training data and rest 20% for validation data.Now, i want to use that validation data to update my trained network as future time-series patterns are more affected by recent time-series data.Any suggestion on how do i make an approach?
You could use EKF to update with new patterns in the hold out set and as they arrive.... This is a difficult one.
Alternatively, use a very small learning rate and retrain the network with the full sample again.
Of course, make backups before you attempt either.
Log in to post a comment.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.