A bug in tool.py #15
Description
Dear @haoxin1998,
Thank you for your valuable benchmark and code. However, I found there may be some questions or bug in the earlystopping part.
1. While the best model is stored in the 'checkpoint.pt', only the 'self.model' part is stored, and other parameters such as 'self.mlp' is discarded.
2. Besides, the default setting in the 'scripts' seems to be different from the paper. For example, the dataset 'Health' has a seq_len of 24, while in the paper the weekly reported dataset is tested on the seq_len of 36; the 'inverse' is False, but the reported metric does not discuss if the metric is normalized.
Therefore, would you please show the original code and setting that generated the reported result? I am looking forward to your reply!