New Metric: Soft Dynamic Time Warping (Soft-DTW)

[VijayVignesh1]

What does this PR do?

Fixes #3084

Before submitting

• Was this discussed/agreed via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

PR review

Added a new domain called timeseries to include Soft DTW loss.
Ref paper: https://arxiv.org/abs/1703.01541

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

---

📚 Documentation preview 📚: https://torchmetrics--3287.org.readthedocs.build/en/3287/

[VijayVignesh1]

@Borda Just wanted to get your initial opinion on this before I move forward. The softDTW uses dynamic programming which makes it really slow to work with on cpus. It's runs faster on gpus. Do you think it's worth continuing the effort?

[Borda]

@Borda Just wanted to get your initial opinion on this before I move forward. The softDTW uses dynamic programming which makes it really slow to work with on cpus. It's runs faster on gpus. Do you think it's worth continuing the effort?

I think it is very fine to have only the GPU version and raise an exception or a warning on the CPU

[justusschock]

let's try to convert it to float? e.g. an int can also be a float and a float-tensor or similar would be fine? Also is there an upper bound?

[VijayVignesh1]

I've removed the float check. Int will be accepted now. Also, there's no upper bound to gamma.

[VijayVignesh1]

@justusschock I'm looking at the unittest results. Turns out it's erroring out:

numba.cuda.cudadrv.error.NvvmSupportError: libNVVM cannot be found. Do `conda install cudatoolkit`:
libnvvm.so: cannot open shared object file: No such file or directory

The pysdtw library used as reference needs it. Is there any way around it?