autoencoder cannot be used

[raqueldias]

When I try to run the summary for a non-convolutional autoencoder architecture:

import torch.nn as nn
import torch
from torch.autograd import Variable
import sys
from torchsummary import summary

class Autoencoder(nn.Module):
    def __init__(self,input_dim, output_dim, n_layers=4, size_ratio=0.5, activation='relu'):
        super(Autoencoder, self).__init__()

        def get_activation(activation):

            if(activation=='relu'):
                return nn.ReLU(True)
            elif(activation=='tanh'):
                return nn.Tanh()
            elif(activation=='sigmoid'):
                return nn.Sigmoid()
            elif(activation=='leakyrelu'):
                return torch.nn.LeakyReLU()

        encoder_layers = []

        in_size_list = [input_dim]
        out_size_list = [output_dim]

        for i in range(int(n_layers/2)):
            out_size_list += [int(out_size_list[i]*size_ratio)]
            encoder_layers += [nn.Linear(in_size_list[i], out_size_list[i+1])]
            encoder_layers += [get_activation(activation)]
            in_size_list += [out_size_list[i+1]]

        decoder_layers = []
        out_size_list.reverse()

        for i in range(int(n_layers/2)-1):
            decoder_layers += [nn.Linear(out_size_list[i], out_size_list[i+1])]
            decoder_layers += [get_activation(activation)]

            decoder_layers += [nn.Linear(out_size_list[-2], output_dim)]
            decoder_layers += [get_activation('sigmoid')]

            self.encoder = nn.Sequential(*encoder_layers)
            self.decoder = nn.Sequential(*decoder_layers)

    def forward(self, x):
        x = self.encoder(x)
        x = self.decoder(x)
        return x

autoencoder = Autoencoder(input_dim=4396, output_dim=4396, n_layers=4, size_ratio=0.5, activation='sigmoid').cuda()
summary(autoencoder, (4396))

I get the error:

Traceback (most recent call last):
  File "test.py", line 54, in <module>
    summary(autoencoder, (4396))
  File "/home/raqueld/.local/lib/python3.6/site-packages/torchsummary/torchsummary.py", line 60, in summary
    x = [torch.rand(2, *in_size).type(dtype) for in_size in input_size]
TypeError: 'int' object is not iterable

Does this package support only CNNs and RNNs? what about feedforward neural networks and autoencoders?

[cainmagi]

Your invocation is not correct. Change the codes like:

summary(autoencoder, (4396,))