Are tables like concattable or paralleltable present in torch

#1

Are there utilities like concattable, paralleltable in pytorch??

1 Like
#2

No, but you can easily achieve that using autograd.

e.g. for concat table:

class MyModel(nn.Module):
    def __init__(self):
        self.submodule1 = MySubmodule1()
        self.submodule2 = MySubmodule2()

    def forward(self, x):
        # this is equivalent to ConcatTable containing both submodules,
        # but is much more readable
        concat = [self.submodule1(x), self.submodule2(x)]
        # do some other processing on concat...
2 Likes
#3

Hi, what if the number of submodule is variable based on the user?

I tried the following

class MyModel(nn.Module):
    def __init__(self, submod_num,submod_var):
        self.submod_pool = []
        for i in range(submod_num):
            self.submod_pool += [MySubmodule(submod_var[i])]

    def forward(self,X):
        outPool = []
        for submod in self.submod_pool:
            outPool += [submod(X)]

        return torch.cat(outPool,dim)

But I’m not sure if it works because the code compile fine but the training error is not as I expected. Then I found tensorboardX, when I used it to draw the graph, it returns an error which clearly indicates that the graph is disconnected

#4

hi @thongnguyen, did you find a solution for variable number of submodules?

#5

@Amir_Ghodrati sorry for late reply, I didn’t notice there was a mention.

I solved it by this:

 model= nn.Sequential(*layer_list)