An offical implementation of PatchTST: "A Time Series is Worth 64 Words: Long-term Forecasting with Transformers." (ICLR 2023) https://arxiv.org/abs/2211.14730
Apache License 2.0
1.51k
stars
262
forks
source link
Question about experimental settings of national illness dataset #76
Hello, I did not find the details of the illness dataset about the look-back window in your paper when I would like to reproduce the PatchTST experiment on the illness dataset. Could you please provide me with the sequence length settings for both models (PatchTST/64 and PatchTST/42) on the illness dataset? I noticed that in the source code provided by your official repository, seq_len is set to 104. I assume this corresponds to PatchTST/42, but I am not sure.
Additionally, for the illness dataset, the patch embedding stride is set to 2, but seq_len is 104. I feel confused because the stride setting does not seem to correspond to either PatchTST/64 or PatchTST/42 naming conventions according your paper, as the number of input patches should be 64 or 42, rather than 52.
Hello, I did not find the details of the illness dataset about the look-back window in your paper when I would like to reproduce the PatchTST experiment on the illness dataset. Could you please provide me with the sequence length settings for both models (PatchTST/64 and PatchTST/42) on the illness dataset? I noticed that in the source code provided by your official repository, seq_len is set to 104. I assume this corresponds to PatchTST/42, but I am not sure.
Additionally, for the illness dataset, the patch embedding stride is set to 2, but seq_len is 104. I feel confused because the stride setting does not seem to correspond to either PatchTST/64 or PatchTST/42 naming conventions according your paper, as the number of input patches should be 64 or 42, rather than 52.