search

KindXiaoming / pykan

Kolmogorov Arnold Networks
MIT License
14.76k stars 1.35k forks source link

symbolic_formula is broken #174

Open awero-manaxiy opened 4 months ago

awero-manaxiy commented 4 months ago

Please specify an error in the except clause!

There is no 'arcsin' function in sympy, it's named 'asin'. auto_symbolic is fixing with arcsin, which causes the program to output 'make sure all activations need to be converted to symbolic formulas first!' even when every activation is symbolic.

Also ex_round is broken as well. With it, code reliably outputs ZeroDivisionError every time on my data. Ended up rewriting the function with torch.round which fixed the issue.

manuelcugliari commented 4 months ago

Same issue (maybe). When I try to get the the symbolic formula (model.symbolic_formula(var=input_features) I get 'ZeroDivisionError: integer division or modulo by zero'

bozhenhhu commented 4 months ago

criterion = F.mse_loss model_kan = KAN(width=[5, 10, 10, 10, 1], grid=50, k=3, seed=args.seed, device=device) print(model_kan.device) results = model_kan.train(temp_dataset, opt="LBFGS", steps=10, lamb=0.01, loss_fn=criterion, device=device, lamb_entropy=2.)

I met the same issue.

AminMoradiXL commented 3 months ago

same issue with arcsin