Implementing one to many LSTM/RNN, PyTorch - pytorch

Question

I have a matrix sized m x n, and want to predict by 1 x n vector (x at the picture with the network structure) the whole next (m-1) x n matrix (y^{i} at the picture), using RNN or LSTM, I don't understand how to implement feeding each
1 x n vector to the next hidden state and get all the
(m-1) x n vectors simultaneously and how to compute error over all y^{i}
I have this vanilla RNN-model and don't know how to modify it
class RNNModel(nn.Module):
def __init__(self, input_dim, hidden_dim, layer_dim, output_dim):
super(RNNModel, self).__init__()
self.hidden_dim = hidden_dim
self.layer_dim = layer_dim
# (batch_dim, seq_dim, feature_dim)
self.RNN = nn.RNN(input_dim, hidden_dim, layer_dim, batch_first=True, nonlinearity='tanh')
self.fc = nn.Linear(hidden_dim, output_dim)
def forward(self, x):
# Initialize hidden state with zeros
h0 = torch.zeros(self.layer_dim, x.size(0), self.hidden_dim).requires_grad_()
out, h_t = self.RNN(x, h0)
#out = self.fc(h_t[:, -1, :])
out = self.fc(out[:, -1, :])
return out

Answer 1

Stainley, try this:
you initiate hidden state only if no other hidden state is passed. then you return the hidden state and pass it to forward() at the next iteration.
def forward(self, x, h=None):
if h is None: # if no hidden state is passed
h = torch.zeros( # Initialize hidden state with zeros
self.layer_dim, x.size(0),
self.hidden_dim).requires_grad_()
out, h_t = self.RNN(x, h)
out = self.fc(out[:, -1, :])
return out, h_t
in training code you run the cycle like this like this:
x = seed
h = None
for i in range (...)
optimizer.zero_grad()
...
x, h = model.forward (x, h)
...
loss = ...
loss.backward()
optimizer.step()