Performance issues that scale with total number of samples to annotate.

[spott]

For many samples (>1000), the lag between when a button is clicked and the display of the next sample gets longer (in a way that is roughly proportional to the number of samples).

I'm not sure what is causing this. It appears that there is a lot of iterating through the entire dataset -- for example, when updating the progress indicator, it iterates through the entire task list, rather than just updating the one value that has been changed.

It would be awesome if there was a way to do the annotation on a generator, rather than coercing everything to a list. I know it would change the progress indication, but it would fix this problem.

[kuk]

It happens because ipyannotate redraws interface on each update. We need to work on javascript side to make it faster. Two reasons why it was not done yet:

1. I am a bad javascript programmer
2. ipyannotate is for manual annotation. It is hard for human to work with >1000 tasks. In practice one splits data in ~100 tasks chunks any way

Supporting annotation of generators is another feature request. Currently not sure why ipyannotate would need that. Fixing performance issues requires javascript implementation optimisation

[spott]

So, I think I'm running into jupyter-widgets/ipywidgets#1345 with this. When I try and look at ~100 tasks, the memory will slowly increase (along with taking longer per iteration).

Do you have a good idea about how to "close" a widget in order to free up the browser space?