High CPU usage

[partofthething]

This component uses a LOT of CPU, especially when running with constellations (but even without). In profiling, we can see that the python-skyfield library computing positions is a lot of work. Here is the profiler trace with constellations:

We could alternatively make a sun-only mode that uses the alternative library pysolar which would likely be much faster.

[brandon-rhodes]

I am planning to write up soon a guide to faster Skyfield calculations, especially in cases where one does not need observatory-telescope accuracy in the positions. But first: may I ask what the frame rate is here? There's likely only a need to compute a position each, say, 10 minutes, and momentary positions could be linearly interpolated between the 10-minute positions with very little loss of precision.

[partofthething]

Awesome! I'm really happy to hear that.

A frame rate of once/10 minutes would be completely adequate for this particular application, as when you're less than that the changes really aren't perceptible on this kind of a plot. In fact, one intermediate work-around would be to throttle the component to that kind of rate. At the moment, my update rate is closer to once/10 seconds, but that's just totally unnecessary.

Still, fast lower-accuracy calcs would be super useful for this as even the once/10 min calc would be a struggle for a Raspberry Pi I think.

[brandon-rhodes]

Roger. I'll see about getting those docs written in the next few days, and I'll comment here with a link!

[brandon-rhodes]

As it's now been a fully quarter since my last comment, I should provide an update: I was derailed for a while on the question of why abbreviated nutation series were not behaving within the bounds I had seen claimed. Having returned a few times and continued my investigation, I think the numbers that folks offer are actually not meant to be true upper bounds on the error — hence the disagreement with the numbers I was getting.

So at some point over this winter I think I'll have time to sit down and put together solid numbers of my own about how Skyfield users can sacrifice accuracy for speed by using less exact nutation computations. I still plan to provide an update here when I have something written up!

[partofthething]

@brandon-rhodes thank you so much for checking in on this.

[roumano]

Actually, the image is redraw every 10secondes.
It's should be good if it's exist a option to increase the delay between refresh.