[Request] A more accurate ETA for jobs with varying timesteps

[jaredjeya]

In Python's TQDM, you can set a "smoothing" parameter which determines how the ETA is calculated. As I understand it, this interpolates between using the average speed and the instantaneous speed, using an exponentially decaying weighting for old values. The relevant code can be found at https://github.com/tqdm/tqdm/blob/0ed5d7f18fa3153834cbac0aa57e8092b217cc16/tqdm/std.py#L214.

Is there any possibility of something like this being implemented, please? My use case for this is that I have a function that takes varying times (in my case, it's slow at the start of the loop but speeds up), so the time taken to finish is very different from that displayed by the ETA.

Even having the choice between instantaneous speed (i.e. time for last iteration) and average speed (i.e. total time/total iterations) would be a big improvement.