if input.dim() == 2 and bias is not None: AttributeError: 'tuple' object has no attribute 'dim'

1 
import rpy2.robjects as robj
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable

robj.r["source"]("input_matrix.R")
matrix = robj.r['input_matrix']
HIDDEN_LAYER = 5
LEARNING_RATE = 0.01
data_train = torch.tensor(matrix)
data_train = data_train.view(10,891)
print(data_train.size())
INP_DIM = data_train.size()[0] - 1

class Titanic(nn.Module) :
        def __init__(self,d_in,h):
            super(Titanic,self).__init__()
            self.layer1 = nn.Linear(d_in,h)
            self.layer2 = nn.Linear(h,h)
            self.layer3 = nn.Linear(h, h)
            self.layer4 = nn.Linear(h,d_in)
            assert(d_in == data_train.size()[0] - 1)
            assert(h == HIDDEN_LAYER)

        def forward(self, *input) :
            input = F.relu(self.layer1(input))
            input = F.relu(self.layer2(input))
            input = F.relu(self.layer3(input))
            input = self.layer4(input)
            return F.log_softmax(input)

net = Titanic(INP_DIM,HIDDEN_LAYER)
print(net)
optimizer = optim.SGD(net.parameters(), lr = LEARNING_RATE)
crit = nn.KLDivLoss()
print(INP_DIM)
for epoch in range(0,data_train.size()[1]) :
        vec = data_train[:,epoch]
        print(vec)
        target = vec[0]
        data = vec[1:]
        print(data)
        target,data = Variable(target),Variable(data)
        optimizer.zero_grad()
        net_out = net(data) #error on this line
2

The error comes from the * in your forward method.
Just remove it of use input = F.relu(self.layer1(input[0])).

May I ask for the reason you are passing your input as *input?

1 Like
3

Thanks that worked.
I was a bit confused on how the input data is to be passed since i am new to pytorch