修改预测序列长度报错

[preetypig]

您好！作者。
我在使用自己的数据集复现您的代码时，首先使用TSformer.py生成pt文件（修改了input_len=247已经num_tokens=247/12），在STEP.py上预测output_len=12时能够顺利复现您的代码，并取得优秀性能！
但改变STEP.py文件中的output_len=48后，出现以下报错：
Traceback (most recent call last):
File "/mnt/3.6TB-DATA/liji/BasicTS-master/experiments/train.py", line 29, in
main()
File "/mnt/3.6TB-DATA/liji/BasicTS-master/experiments/train.py", line 23, in main
basicts.launch_training(args.cfg, args.gpus, node_rank=0)
File "/mnt/3.6TB-DATA/liji/BasicTS-master/basicts/launcher.py", line 118, in launch_training
easytorch.launch_training(cfg=cfg, devices=gpus, node_rank=node_rank)
File "/home/pcz4514/anaconda3/envs/liji_graph/lib/python3.9/site-packages/easytorch/launcher/launcher.py", line 86, in launch_training
train_dist(cfg)
File "/home/pcz4514/anaconda3/envs/liji_graph/lib/python3.9/site-packages/easytorch/launcher/launcher.py", line 35, in training_func
raise e
File "/home/pcz4514/anaconda3/envs/liji_graph/lib/python3.9/site-packages/easytorch/launcher/launcher.py", line 31, in training_func
runner.train(cfg)
File "/mnt/3.6TB-DATA/liji/BasicTS-master/basicts/runners/base_epoch_runner.py", line 430, in train
loss = self.train_iters(epoch, iter_index, data)
File "/mnt/3.6TB-DATA/liji/BasicTS-master/basicts/runners/base_tsf_runner.py", line 345, in train_iters
loss = self.metric_forward(self.loss, forward_return)
File "/mnt/3.6TB-DATA/liji/BasicTS-master/basicts/runners/base_tsf_runner.py", line 321, in metric_forward
metric_item = metric_func(**args)
File "/mnt/3.6TB-DATA/liji/BasicTS-master/baselines/STEP/loss/step_loss.py", line 13, in step_loss
loss_pred = masked_mae(prediction=prediction, target=target, null_val=null_val)
File "/mnt/3.6TB-DATA/liji/BasicTS-master/basicts/metrics/mae.py", line 34, in masked_mae
loss = torch.abs(prediction - target)
RuntimeError: The size of tensor a (37) must match the size of tensor b (5) at non-singleton dimension 3
请问是否是STEP.py中还有相应参数需要改变，您是否可以解答该问题，谢谢！

[preetypig]

对了，补充上面的问题。我的数据集有17个特征，但我只需要其预测最后5个特征，所以我在TSFormer.py和STRP.py中都加入了一行CFG.MODEL.TARGET_TIME_SERIES =[12, 13, 14, 15, 16]的代码。