Just a question, when I try evaluating the multi agent baseline, I get the error:
File "/shared/home/sf08116/workspace/forecast-mae/eval.py", line 19, in main
model = Model.load_from_checkpoint(checkpoint)
File "/shared/home/sf08116/workspace/forecast-mae/eval.py", line 39, in <module>
main()
RuntimeError: Error(s) in loading state_dict for Trainer:
Unexpected key(s) in state_dict: "net.hist_embed.levels.0.blocks.0.attn.rpb", "net.hist_embed.levels.0.blocks.1.attn.rpb", "net.hist_embed.levels.1.blocks.0.attn.rpb", "net.hist_embed.levels.1.blocks.1.attn.rpb", "net.hist_embed.levels.2.blocks.0.attn.rpb", "net.hist_embed.levels.2.blocks.1.attn.rpb".
Adding strict=False when loading the checkpoint bypasses the problem but the metrics are slightly higher than reported:
But the main problem is scoring, the best score is always given to modality 1 (thus the high classification loss). Is this because of the strict=False? Do you also always get modality 1 as the best in the scores?
Hello! Thank you for your work.
Just a question, when I try evaluating the multi agent baseline, I get the error:
Adding
strict=False
when loading the checkpoint bypasses the problem but the metrics are slightly higher than reported:But the main problem is scoring, the best score is always given to modality 1 (thus the high classification loss). Is this because of the
strict=False
? Do you also always get modality 1 as the best in the scores?