added seed for reproducibility for mini-batches.

geffy / tffm

TensorFlow implementation of an arbitrary order Factorization Machine

MIT License

780 stars 176 forks source link

added seed for reproducibility for mini-batches. #12

Closed binga closed 7 years ago

binga commented 7 years ago

This PR ensures the results are reproducible when mini-batches are used during training.

geffy commented 7 years ago

Thanks for contribution! Seems like this fix mini-batch permutation, but initialization still random. Maybe seed needed to be set at tensorflow level.

binga commented 7 years ago

Yes, mini-batch permutation is taken care of. But isn't the tensorflow initialization take care of here: https://github.com/geffy/tffm/blob/master/tffm/base.py#L203. I'm able to reproduce my results with this PR.

geffy commented 7 years ago

Oh, yes, you are right, Alexander already fix this.