Matan Addam

P.S. when not using parallel=True, the speedup is approaching x4 that of pure python under a proper benchmark on my mentioned chipset (for the above number of regression iterations).

Explicitly wiring in numba's `prange` in the function while keeping the `@njit(parallel=True), it simply degresses to no speed up at all v.s. regular python. This is with numpy version 1.24.4,...

Is this issue still alive? Does it mean the http protocol should not be used yet? I'm aware of the alpha nature of this library.

Nice to see initial traction :-))

I would imagine that an OS-specific codec would be more efficient than a python implementation, there seems to be a [thread about such a component for Windows here](https://github.com/microsoft/PowerToys/issues/6087). Maybe there'll...

Negative distances seem to be a possible output of the algorithm, a feature which is addressed in a later work that is also implemented at: https://github.com/google-research/soft-dtw-divergences

Maybe you hit into it also outside of using this particular project. E.g. on my system I get the same error (albeit on a slightly different line of code) by...

The warning message comes from [this line of code](https://github.com/hzbd/kazam/blob/491869ac29860a19254fa8c226f75314a7eee83d/bin/kazam#L75), implying that the python version on 20.04 that you have by default or by manual installation is too advanced for the...

Do you imply many Spark API (features) aren't supported?

Do you think that this represents the installability status for mediapipe in general on Mac M3 (or M1 and higher), or rather confined to Model Maker features?