random seed

[kbrulois]

Wondering what's the best way to get the random seed integer into the pdb file names?
Thanks

[DimaMolod]

Hi, you can set up random_seed with --random_seed flag then each model will have individual random seed
There is no direct way of changing the pdb file names, but you can either add a print line there or calculate it manually. The default value for random seed is 42.

[kbrulois]

Hi, thanks.
A follow up question. If I run num_predictions_per_model=2 with random_seed=42, I'd get 10 model seeds: 420-429.
Then if I wanted to reproduce the model for seed 426 (or get as close to it as possible) but with num_predictions_per_model=1 instead of 2, I would have to set random_seed=85, to have one of the 5 models be run with seed 426.

Is my understanding correct?

[DimaMolod]

Actually, we have just noticed that random seed is determined in AP and this is not an expected behaviour, thanks for bringing this up! I will fix that and add a logging about the random seed later today, so maybe you could just wait a bit.

[DimaMolod]

Hi @kbrulois
I've just committed the fix, now random_seed is randomly generated depending on the backend in use and the value is reported in the logs. Please let me know if you have any issues/suggestions