Break states into smaller states to get higher fidelity pitch

sot / chandra_cmd_states

Interface for creation and maintenance of the Chandra commanded states database.

https://sot.github.io/cmd_states

BSD 3-Clause "New" or "Revised" License

3 stars 1 forks source link

Break states into smaller states to get higher fidelity pitch #21

Closed jeanconn closed 8 years ago

jeanconn commented 9 years ago

Improve pitch value for long dwells by adding mock cmds to "clock out" pitch at intervals.

jeanconn commented 9 years ago

I was just curious if this was a prettier or uglier idea than #13 .

jeanconn commented 8 years ago

@taldcroft I think I like this better overall than #13, and I'd like to go with it. Do you have ideas for what would be sufficient testing?

taldcroft commented 8 years ago

Testing could be:

Generate commanded states for a "long-enough" interval to see everything change, maybe a month.
Use the function which squashes the full commanded states down to selected parameters of interest (I forget the name). Select all parameters except pitch.
Confirm that flight and test give the same results.
Confirm that no intervals in test are longer than the max state time.
For a long dwell confirm that sun pitch is changing as expected.

jeanconn commented 8 years ago

When you say ""long-enough" interval to see everything change", are you talking about just trying to get a reasonable sampling of our variety of cmds and cmd_states to look for any interactions? Which ways would I confirm it was long-enough/sufficient?

taldcroft commented 8 years ago

Yes, just get a sampling of commands. You'll have to use your judgement on what is sufficient.

jeanconn commented 8 years ago

Questions on this lingering PR:

Is it still worth doing? How far away are we from dynamic cmd states?
When run on historical data, do you want the new time-sampled states or the previous behavior? Easier, of course, to just always do the behavior, but I'll need to update the regression test data in the timelines project to match.
For testing, you mentioned "Use the function which squashes the full commanded states down to selected parameters of interest (I forget the name). Select all parameters except pitch.". In my initial go at this, I run into the problem that sampling to get you updated pitch also gives you new data for all of the pointing-related columns. So that part of the test would just need to select on all columns that aren't related to pointing. I think it would also be useful to check both that the new states match the old states within a tolerance for new states that start at the same time as old states.

jeanconn commented 8 years ago

I fixed a bug in 063ac6e and put testing notes in

https://github.com/sot/cmd_states/blob/2f26e493c9e2aaec9bb34765ee17d1cf8a022e08/resample_states_verification/NOTES.md

I think this is good to go at some point.

taldcroft commented 8 years ago

Why is test_minus_point.dat in the repo?

taldcroft commented 8 years ago

Why are there no pitch points after 2010:216:00:00:00 through the end of that dwell?

taldcroft commented 8 years ago

Looking at the plot (very illuminating, glad you included it!), it seems like the interval could be 10000 sec instead of 5000 sec without much practical loss of resolution.

jeanconn commented 8 years ago

Regarding

Why are there no pitch points after 2010:216:00:00:00 through the end of that dwell?

That was due to two different "shortcomings". First, the pitch commands were being made within the range of the times of the actual commands instead of from the range starting with the start of the seed state. So, if there were no actual commands for quite a while after the start state of an update, there'd also be no clock out commands. Also, the reference 'last_date' was set at the first "real" command so any clock out pitch commands would be ignored until the first real command. Both fixed. Will update the plot next.

jeanconn commented 8 years ago

Reran testing after those functional changes in f376d23 and see no changes.

taldcroft commented 8 years ago

:+1: for merge and install.

jeanconn commented 8 years ago

This is installed as Chandra.cmd_states 0.10. There's a backup of cmd_states in SKA/data/cmd_states/cmd_states_8jun2016.h5 and a backup of the sybase tables in SKA/data/sybase_backup if we need to revert for any reason.