-
Notifications
You must be signed in to change notification settings - Fork 814
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Help Reproducing the TiDE Paper #2343
Comments
Hey! There are a few issues I can see at a high level with the replication.
They also use learning rate scheduling in the paper.
|
Hi @alexcolpitts96 Thank you for taking the time to reply! I wasn't sure what to do with the data splits, the paper does say 7:1:2, but their code has 6:2:2, unless I am misinterpreting their code. I tried both and it didn't change the results significantly. I made the other changes and the results were MAE: 0.91 MSE: 1.39, now higher than the reported results. I'll include my updated code. Any other suggestions on what I might be doing wrong? Thank you again!
|
I see that they used gradient clipping. You can use it with darts by passing it to I am also not sure you configured the learning rate scheduler to be identical. I am not super familiar with the settings since I generally don't use cosine annealing. They also have seeds posted ( |
Hi @kevincoakley, you should also inverse transform the historical forecasts predictions and compute the metrics against the un-transformed test series. |
Hi @dennisbader Thank you for taking the time to answer my question. I am not clear what you mean by un-transformed test series. I have been confused if I was evaluating the predictions properly. Do you mean something like this?
|
@kevincoakley, yes that's it :) |
Hi, I am trying to reproduce the TiDE paper (https://arxiv.org/abs/2304.08424) using Darts and I am having some issues.
The paper reports results MSE: 0.454 MAE: 0.465 for the ETTh1 (720) dataset, however I am getting MSE: 0.026 MAE 0.127. Am I doing something wrong with the data or the evaluation? Any pointers would be appreciated. Thank you.
The text was updated successfully, but these errors were encountered: