Multivariate sequence to univariate sequence

[njujinchun]

Hi,

I'd like to use, for example, LSTM for univariate time series (TS_a) forecasting given multivariate time series (including the target ones, e.g., TS_a, TS_b, and TS_c) as inputs. Is this feasible? In lstm_model.py:

    self.encoder = LSTM_Encoder(
        input_dim=time_dim + d_yc,
        hidden_dim=hidden_dim,
        n_layers=n_layers,
        dropout=dropout_p,
    )

    self.decoder = LSTM_Decoder(
        output_dim=d_yt,
        input_dim=time_dim + d_yt,
        hidden_dim=hidden_dim,
        n_layers=n_layers,
        dropout=dropout_p,
    )

It requires that "d_yt == d_yc" (i.e., the number of the target time series should equal to that of the context time series). My question is how to handle the case with d_yt=1 and d_yc=3 in LSTM?

Thanks,

Shaoxing