gumityolcu
gumityolcu
I also need this, especially can someone clarify where the random projection trick is being done ? Thanks
https://numpy.org/doc/stable/reference/random/generator.html Using dedicated RNG will solve this 👍🏼 With correct versions of course. I seem to remember an equivalent class in torch. What I can find is this https://pytorch.org/docs/stable/generated/torch.Generator.html But...
@dilyabareeva my proposal is given in the randomizartion metric. takes a seed from the user and initiates a torch.Generator object to use for randomization. This way, using the same exact...
Another example: Metrics require torch.nn.Module but we allow for Benchmarks to pass Lightning modules to metrics
This is currently handled by sending model to the device every time after training. Hence, i am removing the pre-release label
Different random numbers maybe? Do you mean that the explanations for the same (model, test sample, seed) change between OSs, or do you mean something more?