About running experiments

[pprobst]

Hello.

I've recently read your paper "Neural Network Heuristics for Classical Planning: A Study of Hyperparameter Space" (ECAI 2020) and I tried to replicate some of your results with the code provided with the paper. However, a colleague and I ran into some trouble at building that version of Neural Fast Downward, and after some frustration, we ended up trying the most recent version of Neural Fast Downward available it this repository. Thankfully, with this version of NFD, building it worked perfectly and I was also successful at running the PyTorch example.

However, we're a bit lost and it's still unclear to me how we'd be able to replicate some of your results, at least with the current version of NFD. If it's not much trouble, could you elaborate on how to replicate a small example like e.g Blocks with varying hidden layers as in Figure 3 of your paper?

I'm very new to Fast Downward and planning in general, so excuse me if any of my questions seem obvious.

[pprobst]

After some effort and reading on the new docs, we were able to do what we wanted. So I'm closing this issue.
Thanks for the updated documentation.