Is it possible to retrieve the configurations suggested by meta-learning without fitting the model?

[MartinKnaze]

Hi, I would like to retrieve the configurations suggested by meta-learning without waiting for the fitting to finish.

Following the steps in this issue, it is clear that the meta-learning configurations are initialised within seconds. However, to read them from the log file, I'd need to let the fitting run for at least 30 seconds, since that is the lower limit for the 'time_left_for_this_task' argument.

Is there any way to circumvent the need to wait for 30 seconds and retrieve the meta-learning configurations as soon as they are identified?

Thank you!

[eddiebergman]

Hi @MartinKnaze,

Technically I believe this is possibly but not exposed in any way. I am on my phone and won't be back until next week but I think this could be a great addition to autosklearn. I will reply with more detail once back!

Best,
Eddie

[eddiebergman]

Hi again @MartinKnaze,

This is not currently possible and I will mark this as a feature to implement for the future!

Best,
Eddie

[MartinKnaze]

Hi @eddiebergman,

Thank you for letting me know!

Best,
Martin