When I try to ues 1440 data to predict 1 time-series it shows this error

[ooOPPPPPS]

parser.add_argument('--seq_len', type=int, default=1440, help='input sequence length for global_model')
parser.add_argument('--label_len', type=int, default=10, help='start token length') #720
parser.add_argument('--pred_len', type=int, default=1, help='prediction sequence length')

File "/home/panhaokang/Documents/GCformer-main/layers/SelfAttention_Family.py", line 67, in _prob_QK
M = Q_K_sample.max(-1)[0] - torch.div(Q_K_sample.sum(-1), L_K)
IndexError: max(): Expected reduction dim 2 to have non-zero size.

[YanjunZhao]

The model does not support prediction length=1 because of the attention structure in the decoder.
You can follow the setting in the code and print the first value of the prediction result.

[ooOPPPPPS]

Dear Owner:

what about I want --features is S and set
parser.add_argument('--enc_in', type=int, default=1, help='encoder input size')
parser.add_argument('--enc_raw', type=int, default=1, help='encoder input size')
parser.add_argument('--dec_in', type=int, default=1, help='decoder input size')

it will get this error:

return x[:, :, channel_sample], y[:, :, channel_sample]
IndexError: index 332 is out of bounds for dimension 0 with size 1

Is that also the problem from attention structure?

I took the same setting process in fedformer autoformer informer, all of them can get correct output

Bests,

[YanjunZhao]

Yeah, maybe you can try to modify the decoder part of the code to predict univariate time series.