Closed nilsleh closed 1 year ago
Yes. The default configs specified in the implementation of CuboidTransformerModel
are arbitrary. To reproduce the results shown in the paper, you may want to refer to the cfg.yaml
file of each dataset for the exact hyperparameter configs (E.g., we set enc_depth = [4, 4]
and dec_depth = [4, 4]
by default for experiments on N-body MNIST using the corresponding cfg.yaml
.).
Let me close the issue for now. Please feel free to reopen if you find any problems.
The default arguments for the CuboidTransformerModel include
enc_depth=[4, 4, 4]
anddec_depth=[2, 2]
and during the init the lineassert len(enc_depth) == len(dec_depth)
yields an error.