Our current version restarts from training from initial weights, we should have an option where we gradually retrain the models as new data are acquired.
We'll need to keep track of separate training sets of each model so that we don't get data pollution between training/validation steps across multiple re-training events.
Our current version restarts from training from initial weights, we should have an option where we gradually retrain the models as new data are acquired.
We'll need to keep track of separate training sets of each model so that we don't get data pollution between training/validation steps across multiple re-training events.