Poor performance on Energy dataset

[francishijack]

I can't achieve similar results on the Energy dataset as in the paper.
I use the default setting of Stock: main_stock.py to train on the Energy dataset. The result is that: Discriminative Score = 0.454, Predictive score = 0.269

I think the parameters in main_stock.py might not be suitable for the Energy dataset (high dimensional). So I changed --layers=5 , --embed-dim=144 and made more epochs and iterations in the AE training part. However, little has been achieved.

How can I get ideal results on the Energy dataset?

[acphile]

I use --hidden-dim 112 --embed-dim 448 --noise-dim 448 --layers 3 in practice for Energy dataset. --hidden-dim should be 4 times the size of input features as the paper says and embed_dim=(layers+1)*hidden_dim is a feature of our model.