Create the command line argument --rand-seed to allow the user to set the random seed in PyTorch. This is beneficial because it will facilitate better comparisons between different model types, decreasing the variance error of the model.
In the future, we could look into implementing torch.use_deterministic_algorithms(True) as described in the PyTorch Reproducibility Documentation. For me, however, this causes CUDA errors that would limit its performance, making me tend towards not implementing this functionality unless deemed necessary in the future.
Create the command line argument
--rand-seedto allow the user to set the random seed in PyTorch. This is beneficial because it will facilitate better comparisons between different model types, decreasing the variance error of the model.In the future, we could look into implementing
torch.use_deterministic_algorithms(True)as described in the PyTorch Reproducibility Documentation. For me, however, this causes CUDA errors that would limit its performance, making me tend towards not implementing this functionality unless deemed necessary in the future.