Raphaël Robidas

Thanks for the detailed reply. I understand the source of the issue, but it seems very unpractical. I had never heard of the RAD specification and I don't know a...

By "3D builder", I mean software to construct 3D structures like Gaussview, Avogadro, IQmol, Spartan... Creating inputs for calculations is not the issue. The issue is precisely that these inputs...

I see, I've always considered that everything was explicit unless specified otherwise. That makes sense, although the behaviour of the program is still strange to me: ``` $ obabel ch3.mol...

I don't think that the lack of explicit connectivity is a huge issue here, it can be guessed quite accurately from the interatomic distances. As long as there aren't any...

In absence of charge, why not create the InChI with an undefined charge layer? It's good enough far most uses and that's what the official `inchi` package does. Users that...

Thanks for your very quick reply and patch! Indeed, the path is pretty odd and I didn't necessarily expect Pysisyphus to nicely converge to a transition state, but the exception...

@akensert Thanks for looking into this. Your solution does solve the issue in the example perfectly, thanks a lot! It seems to be problematic for some kinds of layers though....

Actually, that's not quite true. The problem occurs when the input specifications are not defined explicitly in the model via `keras.layers.Input(type_spec=X_train.spec)`. Adding this seems to solve the issue as far...

PBEh-3c is a special case, since it was designed with specific (minimal) basis sets and corrections to offset the shortcomings of these small basis sets. When used, no basis set...

I would be also interested. Si was recently reparameterized for GFN1-xTB (10.1021/acs.jcim.1c01170). I am guessing that GFN1-xTB is easier to reparameterize than GFN2-xTB? That could be a starting point for...