[Feature Request]: Fit without validation

[tsoernes]

Is your feature request related to a problem? Please describe.

After having found optimal hyperparameters through many iterations, I want to fit the model using ALL available data. This is necessary to do with time series forecasting with Prophet in order to actually use the model in production for predicting the future afterwards.

How to do so?

Creating a Prophet library through the native library is not a valid solution since FLAML creates time features automatically behind the scenes etc.

Describe the solution you'd like

An argument to automl.fit or another method where you just have 1 iteration without validation / testing in order to produce a production model using ALL data.

Additional context

No response

[thinkall]

Hi @tsoernes , please check out this doc, you can set retrain_full to True to retrain model on the full training data. And retrieve the best model like this