Scheduler callable interface question

[wtgee]

I don't understand the reasoning behind making a Scheduler class callable instead of just making the make_schedule method public. It doesn't seem very intuitive to me.

Also, as it exists now there is a bug somewhere in _make_schedule such that it cannot actually be called twice in a row. So, for example in the #208 that @kvyh just posted (linked below) there is a separation between the creation of a Scheduler and the calling of it, but you can't actually ever call it twice. (That's probably a separate Issue)

https://github.com/kvyh/astroplan/blob/7cf328a4ebadcc6592cc4c68d63df7e0cf6e88fa/docs/tutorials/scheduling.rst#scheduling

[wtgee]

Actually, you can't call it twice because it is reusing the self.schedule in a weird way. If you recreate self.schedule at the top of _make_callable (instead of in init) then it works twice in a row. However, there is some weird caching going on. The first call takes ~45s on my laptop whereas the second call is 600ms. (%timeit reports the discrepancy)

[bmorris3]

@wtgee You might not know that @kvyh is a Google Summer of Code intern working with fleshing out the first scheduler implementations in astroplan. It will take some time before any of these scheduler features are fully supported, and we appreciate your patience (and help!) in the mean time.

The API choice of making Schedulers callable is so that you can call a scheduler with the same properties on different sets of observing blocks. The current version is not prepared to do that properly yet.

For reference/discussion on the API decisions, take a look at #178 and the discussion there.

[kvyh]

When I was writing the code, I wasn't considering using the same scheduler twice. Moving the creation of the schedule to the _make_schedule should allow it to be called multiple times, I will fix that as soon as I have time. Some constraints do cache values, which has the potential for speeding it up the second time.

[wtgee]

Yeah, no worries and I didn't mean to put any pressure on or call anything into question, I was just wondering if there was a design decision around the callable. I don't quite understand the response but I have not good reason to change it either. :)

And I know that it's not being optimized for speed or anything at this point. I realize that just getting it working is a ton of effort and appreciate the work @kvyh is doing. If I'm complicating things by jumping into the code too early let me know!

[wtgee]

@kvyh I guess my thought is you would have one type of Scheduler that you would use to make multiple Schedules by passing different ObservingBlocks. We have long running processes that would need to generate a new Schedule every night, ideally from the same Scheduler that was created when the process was created. That could be changed depending on the intended use however.

[kvyh]

The utility of being able to call the same Scheduler multiple times is great, I'll be doing a PR soon with a few changes talked about in #217. That should make it possible to use the same Scheduler multiple times, as long as you have a new Schedule to input. I hope this makes things simpler and helps clear any confusion.