Autoencoder_RuntimeError: shape '[-1, 784]' is invalid for input of size 20

deployment
1

Hi everyone, just recently started to implement an autoencoder model using pytorch. My data is cleaned and ready to get worked with. For the start, I followed a tutorial on autoencoders in pytorch and could not find out where the error is coming from. It’d be great if anyone could help!

data_loader = torch.utils.data.DataLoader(dataset = training_data, batch_size = 10, shuffle = False)

class Autoencoder_Linear(nn.Module):
def init(self):
super().init()
self.encoder = nn.Sequential( # we use sequential model, where we repeatedly apply linear layers
nn.Linear(28 * 28, 128), # (N, 784) → (N, 128)
nn.ReLU(),
nn.Linear(128, 64),
nn.ReLU(),
nn.Linear(64, 36),
nn.ReLU(),
nn.Linear(36, 18),
nn.ReLU(),
nn.Linear(18, 9)
)

    self.decoder = nn.Sequential(
        nn.Linear(9, 18),
        nn.ReLU(),
        nn.Linear(18, 36),
        nn.ReLU(),
        nn.Linear(36, 64),
        nn.ReLU(),
        nn.Linear(64, 128),
        nn.ReLU(),
        nn.Linear(128, 28 * 28),
        nn.Softmax() 
       # Softmax is used as the final layer for multi-class classification problems. 
       # It is used to predict the probability of each output class.
    )

def forward(self, x): # implement forward
    encoded = self.encoder(x)
    decoded = self.decoder(encoded)
    return decoded

`model = Autoencoder_Linear()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(),
lr=1e-1, # learning rate, here: .001
weight_decay=1e-8)

num_epochs = 10
outputs = []
losses = []
for epoch in range(num_epochs):
for (city, _) in data_loader:
# reshaping the city
city = city.reshape(-1, 28*28)
reconstructed = model(city) # reconstructed model
loss = criterion(reconstructed, city) # criterion with reconstructted cities and original cities

    optimizer.zero_grad() # zero gradients
    loss.backward()
    optimizer.step()

print(f'Epoch:{epoch+1}, Loss:{loss.item():.4f}') # print epoch and loss after epoch
losses.append(loss)
outputs.append(epoch, city, reconstructed) # store in outputs list

`
Output:

RuntimeError Traceback (most recent call last)
in
5 for (city, _) in data_loader:
6 # reshaping the city
----> 7 city = city.reshape(-1, 28*28)
8 reconstructed = model(city) # reconstructed model
9 loss = criterion(reconstructed, city) # criterion with reconstructted cities and original cities

RuntimeError: shape ‘[-1, 784]’ is invalid for input of size 20

2

Based on the error message it seems your input data has 20 elements while you are expecting samples with 784 features. If your inputs are expected to have 20 features only, change the in_features value of your first linear layer to 20 too.

1 Like