automl / TabPFN

Official implementation of the TabPFN paper (https://arxiv.org/abs/2207.01848) and the tabpfn package.
http://priorlabs.ai
Apache License 2.0
1.22k stars 109 forks source link

silly question: learning curves #27

Closed amueller closed 1 year ago

amueller commented 1 year ago

This is probably a stupid question, but I was wondering if/how you stored learning curves during training. I can't find anything in the code that records the losses. Maybe there's some built-in torch mechanism that I'm unaware of?

Obviously I can build a list myself and serialize that but I figure you must have done it some way as well...

SamuelGabriel commented 1 year ago

Actually, we never really had anything for this... 😅

I use this code on the stdout:

def get_all_losses(stdout):
    return [float(v) for v in re.findall('mean loss (.*) \|',stdout)]

but yeah it is horribly hacky.