search

ardigen / MAT

The official implementation of the Molecule Attention Transformer.
MIT License
234 stars 57 forks source link

How to run Training loop using random search? #6

Closed Jh-SYSU closed 4 years ago

Jh-SYSU commented 4 years ago

Great Work!

I'm trying to repeat the results in your paper, but I'm having trobule using random search to obtain the best results in each dataset.

As you mentioned in the paper, " we extensively tune their hyperparameters using random search", & "We run two sets of experiments with budget of 150 and 500 evaluations". May I ask how you tune this hyperparameters using random search and how to control the budget of 150 and 500 combinations. I have tried to utilize the skorch package to solve this problem but it failed.

Thanks a lot!

Mazzza commented 4 years ago

Thank You!

Number of possible hyperparameter settings is very large. If you try to fit the full grid in memory first and then sample from it, you probably fail due to the full RAM. Skorch could failed because of these memory issues. We recommend to sample hyperparameters on the fly - it's more memory efficient.

We wrote custom module to create the random search. However we will not share this code at the moment.