search

wouterkool / attention-learn-to-route

Attention based model for learning to solve different routing problems
MIT License
1.04k stars 337 forks source link

fixed issue with alpha value and loss combination in warmup baseline #56

Closed LTluttmann closed 5 months ago

LTluttmann commented 1 year ago

fixes #51

simply updating the alpha value inside the if epoch < self.n_epochsstatement suffices to ensure self.alpha<=1 all the time.

Also, fixes the combination of baseline losses as described here

wouterkool commented 5 months ago

Thanks a lot, I think this is how I intended it! However I really don't understand as the code seems to have worked for me (and many others) previously.