I have SCADA data (temporal data) for four vaiables and I want to o a forecasting. So I decided to combine a 2D conv layers to extract data features and then with these features use a LSTM to find a temporal information and make a prediction.
For the convolutional data I am creating a 12X12X4 matrix (because in my problem 144 samples are one day and I want to predict the nex sample). The number of channels is four because I have four variables.
After the Conv2D I am using a LSTM because I want to extract temporal characteristis. I have an error but I dont know why. Can u help me? I share the network model and the train fuctions
define the NN architecture
class HybridNetwork(nn.Module):
def init(self, n_hidden=128, n_layers=2, drop_prob=0.5, lr=0.001):
super(HybridNetwork, self).init()
self.drop_prob = drop_prob
self.n_layers = n_layers
self.n_hidden = n_hidden
self.lr = lr
# convolutional layer (sees 12x12x4 image tensor)
self.conv1 = nn.Conv2d(4, 32, 3, padding=1)
# max pooling layer
self.pool = nn.MaxPool2d(2, 2)
# linear layer (32 * 6 * 6 -> 64)
self.fc = nn.Linear(32 * 6 * 6, 64)
## TODO: define the LSTM
self.lstm = nn.LSTM(64, n_hidden, n_layers, dropout=drop_prob, batch_first=True)
## TODO: define a dropout layer
self.dropout = nn.Dropout(drop_prob)
self.fc2 = nn.Linear(n_hidden, 1)
def forward(self, x, hidden):
#Ingresa imagen 4X12X12
x = self.pool(F.relu(self.conv1(x)))#6X6
#Se tiene imagen 64X6X6
x = x.view(-1, 32 * 6 * 6)
x = self.fc(x)
r_output, hidden = self.lstm(x, hidden)
out = self.dropout(r_output)
out = out.contiguous().view(-1, self.n_hidden)
out = self.fc(out)
return out, hidden
def init_hidden(self, batch_size):
''' Initializes hidden state '''
# Create two new tensors with sizes n_layers x batch_size x n_hidden,
# initialized to zero, for hidden state and cell state of LSTM
weight = next(self.parameters()).data
if (train_on_gpu):
hidden = (weight.new(self.n_layers, batch_size, self.n_hidden).zero_().cuda(),
weight.new(self.n_layers, batch_size, self.n_hidden).zero_().cuda())
else:
hidden = (weight.new(self.n_layers, batch_size, self.n_hidden).zero_(),
weight.new(self.n_layers, batch_size, self.n_hidden).zero_())
return hidden
initialize the NN
model = HybridNetwork()
print(model)
move tensors to GPU if CUDA is available
if train_on_gpu:
model.cuda()
And the train fuction is:
n_epochs = 2000
valid_loss_min = np.Inf # track change in validation loss
train_losses, test_losses, accuracy_losses = [], [], []
for epoch in range(1, n_epochs+1):
# keep track of training and validation loss
train_loss = 0.0
valid_loss = 0.0
h = model.init_hidden(batch_size)
val_h = model.init_hidden(batch_size)
###################
# train the model #
###################
model.train()
for data, target in train_loader:
# move tensors to GPU if CUDA is available
if train_on_gpu:
images, target = data.cuda(), target.cuda()
# Creating new variables for the hidden state, otherwise
# we'd backprop through the entire training history
h = tuple([each.data for each in h])
# clear the gradients of all optimized variables
optimizer.zero_grad()
# forward pass: compute predicted outputs by passing inputs to the model
output, h = model(images, h)
# calculate the batch loss
loss = criterion(output, target)
# backward pass: compute gradient of the loss with respect to model parameters
loss.backward()
# `clip_grad_norm` helps prevent the exploding gradient problem in RNNs / LSTMs.
nn.utils.clip_grad_norm_(model.parameters(), clip)
# perform a single optimization step (parameter update)
optimizer.step()
# update training loss
train_loss += loss.item()*images.size(0)
######################
# validate the model #
######################
model.eval()
for data, target in valid_loader:
# move tensors to GPU if CUDA is available
if train_on_gpu:
images, target = data.cuda(), target.cuda()
# Creating new variables for the hidden state, otherwise
# we'd backprop through the entire training history
val_h = tuple([each.data for each in val_h])
# forward pass: compute predicted outputs by passing inputs to the model
output, val_h = model(images, val_h)
# calculate the batch loss
loss = criterion(output)
# update average validation loss
valid_loss += loss.item()*data.size(0)
# calculate average losses
train_loss = train_loss/len(train_loader.sampler)
valid_loss = valid_loss/len(valid_loader.sampler)
train_losses.append(train_loss)
test_losses.append(valid_loss)
# print training/validation statistics
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}'.format(
epoch, train_loss, valid_loss))
# save model if validation loss has decreased
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(model.state_dict(), 'data_b8_lr000.pt')
valid_loss_min = valid_loss
And I have the following error:
RuntimeError: input must have 3 dimensions, got 2