Dear Authors,

I would like to ask is there a available settings or possibility that CUTS_Plus could be trained to without causal groundtruth graphs. So in this cases we can save the graphs generated by the model, and use svm to classify like for gender classification?

Thanks a lot,

Can you provide yaml configuration files related to the "netsim" experiment in CUTS?

May I ask which 10 individuals in the Netsim experiment have specific data in the dataset? The results obtained by each individual seem to differ significantly.

May I sask if it is possible that the output is not a DAG?
If that is the case, how can we handle it or can we consider it as a causal graph?
Thanks!

Dear @jarrycyx

Could you please share the yaml configuration file for "netsim" experiment in CUTS-PLUS?

Because i'm facing the constant issue below, and i can't figure out why and how to resolve it.

`
graph = cuts_plus.main(data, mask, true_cm, sota_opt.cuts_plus, log, device=device)

UNN-main/CUTS_Plus/cuts_plus.py", line 395, in main Graph = multicad.train(data, mask, data, true_cm)

UNN-main/CUTS_Plus/cuts_plus.py", line 305, in train y_pred, loss = self.latent_data_pred(x, y, mask_x, mask_y)

UNN-main/CUTS_Plus/cuts_plus.py", line 162, in latent_data_pred graph_sampled = sample_bernoulli(Graph, self.args.batch_size)

UNN-main/CUTS_Plus/cuts_plus.py", line 147, in sample_bernoulli return torch.bernoulli(sample_matrix).float()

RuntimeError: Expected p_in >= 0 && p_in <= 1 to be true, but got false. (Could this error message be improved? If so, please report an enhancement request to PyTorch.)
`
Thanks in advance!!

Hello, the code you provided to calculate TN seems to be incorrect. You used ' np.mean((1-causal_graph) * (1-causal_graph)) ' instead of ' np.mean((1-causal_graph) * (1-true_cm)) ', which should be used to calculate TN. (Refer to line-43 in misc.py) .