clarification regarding Module:getParameters()

[Yossi Biton]

i'm using getParameters() in order to train using optim.sgd. 

(btw, i have considered using nn.Optim from fbnn instead, but didn't love the idea of switching from the master branch of nn).

now, to my question...

from the documentation of nn.Module (https://github.com/torch/nn/blob/master/doc/module.md#flatparameters-flatgradparameters-getparameters):

Since the storage of every weight and gradWeight is changed, this function should be called only once on a given network.

 

when i start training getParameters is called once. 

in some cases i want to stop my training and start again with the same model - in that case getParameters is called again for the same model (but not in the same run).

model = torch.load('model.net')

params,gradParams = model:getParameters()

is it ok or something i should avoid ?

[soumith]

That should be okay.

--
You received this message because you are subscribed to the Google Groups "torch7" group.
To unsubscribe from this group and stop receiving emails from it, send an email to torch7+un...@googlegroups.com.
To post to this group, send email to tor...@googlegroups.com.
Visit this group at http://groups.google.com/group/torch7.
For more options, visit https://groups.google.com/d/optout.

[Yossi Biton]

Could you explain why is it ok ?
Based on the documentation you shouldnt call it twice with the same model.

You received this message because you are subscribed to a topic in the Google Groups "torch7" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/torch7/bqbvIKjG4PU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to torch7+un...@googlegroups.com.

[soumith]

If you call it twice within the same model, the handles wrt the first call become invalid. This is not obvious to people usually.

For eaxmple:

x1,dx1 = model:getParameters()

x2,dx2 = model:getParameters()

Now only x2 and dx2 are valid.

x1 and dx1 are stale handles.

[ama...@gmail.com]

In what order are the parameters returned by calling getParameters() on a given module, could you please explain

In my case i need to set the forget gate bias of FastLSTM to 1 ,for that i need to know the order of the parameters