Peter Eastman
Peter Eastman
As I work through the code, I'm finding some differences between how it's described in the paper and how it's implemented. I assume I should match the behavior of the...
Thanks! I'm a bit confused about the role of the lower cutoff. The `CosineCutoff` function goes to 1 at the lower cutoff distance, and then abruptly jumps to zero. Doesn't...
Perhaps I was misreading the code. Is there a reference describing how the lower cutoff is supposed to be implemented? The paper only describes an upper cutoff.
Thanks!
Out of curiosity, when you said it sometimes helps to set a lower cutoff, what's the reason for that? Does it actually come from changing the functional form? Or is...
In that case, just initializing the parameters so you don't put RBFs at short distances should be sufficient. That would save computation time (the version with upper and lower cutoffs...
Any thoughts on this question? Is there actually a reason to change the functional form, or is it really just a matter of not putting RBFs at short distances? I...
It sounds like you're saying the lower cutoff should only be applied for certain purposes, not others? Here are all the ways it gets used in the ET model: -...
> Poorly sampled regions, like close contacts in MD, can be handled by a more robust but less accurate prior model. Perhaps a way of handling that would be to...
You've structured the kernel so that every thread computes only a single interaction: ```c++ const int32_t index = blockIdx.x * blockDim.x + threadIdx.x; if (index >= num_all_pairs) return; int32_t row...