Run customized Starfish star_mix.py jobs

[gully]

Lots of spot-checking needed.

[gully]

First guess at Anon1 IGRINS spectrum modeled with two-component photospheres.

This is before running any spectral inference. A good first guess helps accelerate convergence.

[gully]

A while ago, when I gave a talk at the Bay Area Exoplanet Meeting, Jack Lissauer objected that 50:50 solutions are suspicious-- these mixtures offer the most constraining power. I disagreed, and showed many spectral orders with conspicuous features. Still, it's worth keeping this concern in mind for future prose.

[kevincovey]

Very cool figure! If you are planning to make more of these figures (one for each order of the fit?) would it be possible to adjust the legend so that the labels for the hot and cool components include the best fit temperatures and weights?

And thinking about the weights -- at this stage, I assume starfish is just giving the flux ratio of the contributions of each component, so a 'post-processing' interpretation step would be required to get to the hot/cool area coverage fraction?

Anyway, very cool!

-kevin

On Wed, Aug 30, 2017 at 1:29 PM, gully notifications@github.com wrote:

A while ago, when I gave a talk at the Bay Area Exoplanet Meeting, Jack Lissauer objected that 50:50 solutions are suspicious-- these mixtures offer the most constraining power. I disagreed, and showed many spectral orders with conspicuous features. Still, it's worth keeping this concern in mind for future prose.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/BrownDwarf/xveganx/issues/10#issuecomment-326109466, or mute the thread https://github.com/notifications/unsubscribe-auth/AFM17yB92Lb2Fp2xGT9Y8-8z1Lni15OLks5sdcYYgaJpZM4Oz94Q .

-- Kevin R. Covey Assistant Professor Western Washington University http://myweb.facstaff.wwu.edu/~coveyk/

[gully]

Currently attempting to run on the WWU cluster:

g_gullysm at fe1 in ~/GitHub/xveganx/sf/V827Tau/m103/output/mix_emcee/run01 on master [?]
$ condor_submit run.sub
Submitting job(s).
1 job(s) submitted to cluster 59706.

Had some weirdness with my environment variables not getting carried over from my bash session. Debugging:

g_gullysm at fe1 in ~/GitHub/xveganx/sf/V827Tau/m103/output/mix_emcee/run01 on master [?]
$ condor_submit -interactive
Submitting job(s).
1 job(s) submitted to cluster 59708.
Waiting for job to start...
Welcome to slot1_15@c-0-4!
You will be logged out after 7200 seconds of inactivity.

[gully]

Ok, I finally have the code working on HTCondor. The HTCondor philosophy is different than what I'm used to with other schedulers. There is a notion of "universes", and local files needed for running the job need to be explicitly specified.

$ condor_q

-- Schedd: fe1 : <140.160.143.130:9618?... @ 11/30/17 16:34:30
OWNER     BATCH_NAME     SUBMITTED   DONE   RUN    IDLE  TOTAL JOB_IDS
g_gullysm CMD: run.sh  11/30 16:33      _      1      _      1 60057.0

1 jobs; 0 completed, 0 removed, 0 idle, 1 running, 0 held, 0 suspended

g_gullysm at fe1 in ~/GitHub/xveganx/sf/V827Tau/m103/output/mix_emcee/run01 on master [?]
$ condor_ssh_to_job 60057.0
Welcome to slot1_30@c-2-1!
Your condor job is running with pid(s) 9223.

Here is the submit file:

# Unix submit description file
# sleep.sub -- simple sleep job

executable              = run.sh
log                     = sleep.log
output                  = outfile.txt
error                   = errors.txt
should_transfer_files   = YES
when_to_transfer_output = ON_EXIT
getenv                  = True
universe                = vanilla
initialdir              = /cluster/home/g_gullysm/GitHub/xveganx/sf/V827Tau/m103/output/mix_emcee/run01
transfer_input_files    = config.yaml,s0_o0phi.json
queue

From the ssh you can use htop to inspect the performance. It seemed like other users were also using these computing resources based on the htop, which is also different behavior than I expected. The code actually gets run in a directory called /var/lib/condor/execute/dir_9219, not the local directory. Naively I would think this screws up the operation of Starfish, but actually it could be fine since I use environment variables and absolute paths to specify the data and library paths in the config.yaml file. cool!

/var/lib/condor/execute/dir_9219
keeping grid as is
Using the user defined prior in $xveganx/sf/V827Tau/m103/user_prior.py
2017 Nov 30, 4:35 PM: 1/10 = 10.0%
2017 Nov 30, 4:36 PM: 3/10 = 30.0%
2017 Nov 30, 4:37 PM: 5/10 = 50.0%
2017 Nov 30, 4:38 PM: 7/10 = 70.0%
2017 Nov 30, 4:40 PM: 9/10 = 90.0%
The end.

[gully]

This is done! We ran 22 spectral orders on the H-band spectrum of V827 Tau. We used the ACCRE system at Vanderbilt-- thanks @garrettsomers !

And holy moly the spectra so far look amazing!!

[gully]

[gully]

Check it out @kevincovey

[gully]

This is done for V827 Tau H-band only!