Charles Jekel

I need to respond to https://github.com/cjekel/similarity_measures/pull/29#issuecomment-1374605492 Curious what happens to the cythonization if everything is written in cython, even the distance functions.

Hey! This is great! and totally missed my radar. I promise to look more in depth with this around the time I push in the other PR for the bugfix...

So, maybe it makes sense to fork this repo and go all out on performance. Honestly this library isn't that much to maintain (as there are not many commits), that...

- If I was doing this problem.... I would probably just use MSE and MAE to compute the error on the path of the next 50 data points between GT...

Are partial matches (between your prediction and GT) a good or bad thing in this case? I can quickly write a pytorch variant of DTW that you can try it...

So this is a DTW distance in pytorch ```python def dtwtorch(exp_data, num_data): r""" Compute the Dynamic Time Warping distance. exp_data : torch tensor N x P num_data : torch tensor...

Here is an example of minimizing DTW to fit a model in pytorch https://github.com/cjekel/similarity_measures/blob/dtwtorch/torch_DTW_demo.ipynb The resulting fit is actually pretty good, although it is not cheap to compute the DTW...

I was asking about the shapes and sizes because I was curious if you wanted to match more than one center line prediction at a time (like mini-batching). So I...

It could also be interesting to port the frechet distance to pytorch as well.

I think it will look something like this ```python def dftorch(exp_data, num_data): r""" Compute the discrete frechet distance. exp_data : torch tensor N x P num_data : torch tensor M...