Why Wci, Wcf, Wco are Variables rather than nn.Parameters about convolutional_lstm_pytorch HOT 3 OPEN

import torch
from torch import nn

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.w = torch.rand(2, 2)
        self.w.requires_grad = True

    def forward(self, x):
        return self.w*x

net = Net()

net(torch.rand(2, 2))

tensor([[0.0726, 0.2216],
        [0.3548, 0.8168]], grad_fn=<MulBackward0>)

list(net.parameters())

[]

net.w.requires_grad

True

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.w = nn.Parameter(torch.rand(2, 2))

    def forward(self, x):
        return self.w*x

net = Net()
list(net.parameters())

[Parameter containing:
 tensor([[0.8596, 0.7057],
         [0.6475, 0.5657]], requires_grad=True)]

from convolutional_lstm_pytorch.

You are right, plz refer to convLSTM.py#L14 which states the reason.

from convolutional_lstm_pytorch.

Thank you for your comment. I am new to pytorch and your explanation is not very clear for me, can you elaborate on that? To my knowledge nn.Parameters are included in model.parameters() so that they are updated with each call of optimizer.step() (what do you mean by forever?) I suppose that Wci,Wcf,Wco also should be trainable in the network.

Paper for convlstm : https://arxiv.org/abs/1506.04214
and your explaination,

-if using requires_grad flag, torch.save will not save parameters in deed although it may be updated every epoch.
-Howerver, if you use declare an optimizer like Adam(model.parameters()),
-parameters will not be updated forever.

Thank you

from convolutional_lstm_pytorch.