Hey everyone,
I am currently working on a lstm which should predict a multidimensional output given any input.
To clarify what I mean, I have the following setup:
Input: [batch_size, seq_len, n_features] = [32, 16, 2]
LSTM:
def __init__(self, in_features: int, n_hidden: int, out_features: int, num_layers: int):
super(LSTMModel, self).__init__()
self.in_features = in_features
self.n_hidden = n_hidden
self.out_features = out_features
self.num_layers = num_layers
self.lstm1 = nn.LSTM(
input_size=in_features,
hidden_size=n_hidden,
num_layers=num_layers,
batch_first=True,
bias=True,
)
self.linear = nn.Linear(
in_features=self.n_hidden,
out_features=self.out_features,
bias=True
)
def forward(self, x: torch.FloatTensor) -> torch.Tensor:
batch_size = x.size(0)
h_t = torch.zeros(self.num_layers, batch_size, self.n_hidden, device=x.device).float()
c_t = torch.zeros(self.num_layers, batch_size, self.n_hidden, device=x.device).float()
x, (h_t, c_t) = self.lstm1(x, (h_t, c_t))
x = self.linear(x[:, -1, :])
return x
The Output should be:
[batch_size, target_seq_len, target_n_features] = [32, 4, 2]
The reason for this setup: I want to ‘predict target_seq_len’, where every target is a vector with ‘target_n_features’ given a history (sequence) with the setting (desc. above)
So I made a multidimensional forecasting… But the problem is, that the code I use for this task only gives me an output of [batch_size, target_seq_len]. What can I do to solve the problem?
I use standard PyTorch and pytorch-lighnting for my problem.
Here is also a screenshot of the shapes (console): The first line is the y_pred and the second is the y_true
Do you have any ideas for code-changes or solutions?
~ Linus
