Random initialization of untrained models

[jessicaloke]

Is it possible to set random seeds for the untrained models? How would I go about doing so?

[Oleg Sémery]

Since the initialization of the network weights is a very sensitive thing for quality learning, initialization for all frameworks, except PyTorch, is your area of responsibility. Accordingly, depending on the framework, this is done in different ways.

[jessicaloke]

I'm doing this in PyTorch - how would I do so then?

[Oleg Sémery]

Then weights are already randomly initialized. If you are interested in learning reproducibility, then pay attention to the function torch.manual_seed().

[jessicaloke]

I see; I had use this function but i had thought that there could be a different way of initializing models.
Thank you for the quick response! (: