Can the evaluation data in the paper be reproduced?

#2
by AllennZ - opened

I am also using GraphCast.
I used the dataset of dm_graphcast and used normalized data.
For the final predicted data, after calculation by RMSE, there is a certain gap with the paper's value, which is relatively obvious.
Have you reproduced the data in the paper?

The files in this repo are a mirror of the official weights here: https://console.cloud.google.com/storage/browser/dm_graphcast. You'd have to ask the original authors if you'd like an answer to that question.

Graphcast repo: https://github.com/google-deepmind/graphcast
Corresponding authors:

  • remilam [аt] google [dоt] com
  • alvarosg [аt] google [dоt] com
  • matthjw [аt] google [dоt] com
  • peterbattaglia [аt] google [dоt] com
shermansiu changed discussion status to closed

Sign up or log in to comment