Use of delay outside of progressive_val_score

[cdeterman]

The use of the delay argument within the progressive_val_score function is very useful, however, what if I need to do any additional processing, storing multiple metrics, or returning the model itself (or is the model updating in-place when it is passed)? Is there a way to implement delay directly in the normal loop with predict_one/learn_one?

[MaxHalford]

Hey there. I'm not sure this fully answers your question, but under the hood progressive_val_score is using stream.simulate_qa. The latter should give you more control over the fit/predict loop.