Load settings from config file and use setattr for torch modules?

1

so i just want to create a config file, say, optimizer.config, like:

name: SGD
lr: 0.1
momentum:0.9

and i initialize my optimizer as below:

optimizer = getattr(torch.optim, config['name'])(filter(lambda p: p.requires_grad, model.parameters()),
                                                 lr=config.getfloat('lr')
        )

i don’t want to directly set the flag

momentum=config.getfloat('momentum')

because i may have more parameters and i want to put them in a loop to load.
but when i try to assign value to “momentum”, I found out seems i cannot use setattr:

setattr(optimizer, 'momentum', 0.9)

it does not work…is there any way that i can load settings from a config file for torch modules?

really thanks

2

Can you please mention if there is an error, or what is the behavior.
Also, please try:

momentum = getattr(optimizer, 'momentum')
momentum = 0.9

I am new to PyTorch, but I presume if the __eq__ operator is custom defined by PyTorch to do more than just assigning, if it also edits some computation graph elements, this should work.

3

no error, just cannot assign the value.
your solution is also not working.