Details

Type: Story

Status: Done

Resolution: Done

Fix Version/s: None

Component/s: Data Release Production

Labels:None

Story Points:7

Epic Link:

Sprint:DRP S185, DRP S186, DRP F181

Team:Data Release Production
Description
I ran multiBandDriver.py on HSCR 9813 3,4 to determine if any cModel configuration parameters can/should be changed to improve the running time and/or bestfit model parameters. My conclusions for the tested parameters are:
initial.nComponents (default 3): For 2, initial fits run in 0.67x the time, but the exp/dev fits run more slowly, so the total fit time is 0.94x. Initial fits are slightly worse on average (with large scatter) and a sizeable fraction of the dev/exp fits are significantly worse (deltalnP < 10).
For nComponents=6, the initial fitting time is more than twice as long, with modest increases in the exp/dev lnP (<1 for most galaxies, with scatter).
Recommendation: leave it at 3 or lower to 2.
initial.optimizer.gradientThreshold (default 1e2): 1e3 runs very slightly faster than default and very slightly improves fits. 1e1 is slightly slower and also results in slightly worse initial/final fits.
Recommendation: lower to 1e3.
initial.optimizer.maxInnerIterations (default 20): Zero impact on fits, running time for 8 is virtually identical; 50 is 1% slower.
Recommendation: leave at 20.
initial.optimizer.noSR1Term (default False): True radically changes fitting times by factors of (0.33, 4.37, 4.98)x for (initial, exp, dev) i.e. the initial fit is faster but the subsequent exp/dev are much slower.
Confusingly, the initial fits are better on average, whereas the final dev/exp are worse  this may warrant followup. Many of the very slow fits have faint PSF mags and may be pathological cases.
Recommendation: leave as False.
region.nFitRadiiMax (default 3): Surprisingly, nfitrad=2 initial fits are 0.5% slower than the default, whereas nfitrad=5 is 2% faster. However, the total fitting time scales by a factor of (0.86, 1.38) for the smaller/larger size.
Smaller fitting regions naturally have larger likelihoods, but I can't make a meaningful comparison between likelihoods  presumably the smaller region fits better in the common are and worse outside.
Recommendation: leave as is pending further investigation.
One common feature is that there is a nonnegligible fraction of objects (usually with faint mag_psf) with very small differences in initial fit likelihood but large ones in the dev/exp fit, which should not occur frequently.
The quoted figures can be generated by the notebook created in DM14118 (lsstdev:/home/dtaranu/src/mine/taranu_lsst/cModelConfigs.ipynb).
Attachments
Issue Links
Activity
Field  Original Value  New Value 

Epic Link 

Risk Score  0 
Status  To Do [ 10001 ]  In Progress [ 3 ] 
Sprint  DRP S185 [ 707 ]  DRP S185, DRP S186 [ 707, 727 ] 
Description 
I ran multiBandDriver.py on HSCR 9813 3,4 to determine if any cModel configuration parameters can/should be changed to improve the running time and/or bestfit model parameters. My conclusions for the tested parameters are:
initial.nComponents (default 3): For 2, initial fits run in 0.67x the time, but the exp/dev fits run more slowly, so the total fit time is 0.94x. Initial fits are slightly worse on average (with large scatter) and a sizeable fraction of the dev/exp fits are significantly worse (deltalnP < 10). For nComponents=6, the initial fitting time is more than twice as long, with modest increases in the exp/dev lnP (<1 for most galaxies, with scatter). Recommendation: leave it at 3 or lower to 2. initial.optimizer.gradientThreshold (default 1e2): 1e3 runs very slightly faster than default and very slightly improves fits. 1e1 is slightly slower and also results in slightly worse initial/final fits. Recommendation: lower to 1e3. initial.optimizer.maxInnerIterations (default 20): Zero impact on fits, running time for 8 is virtually identical; 50 is 1% slower. Recommendation: leave at 20. initial.optimizer.noSR1Term (default False): True radically changes fitting times by factors of (0.33, 4.37, 4.98)x for (initial, exp, dev) i.e. the initial fit is faster but the subsequent exp/dev are much slower. Confusingly, the initial fits are better on average, whereas the final dev/exp are worse  this may warrant followup. Many of the very slow fits have faint PSF mags and may be pathological cases. Recommendation: leave as False. region.nFitRadiiMax (default 3): Surprisingly, nfitrad=2 initial fits are 0.5% slower than the default, whereas nfitrad=5 is 2% faster. However, the total fitting time scales by a factor of (0.86, 1.38) for the smaller/larger size. Smaller fitting regions naturally have larger likelihoods, but I can't make a meaningful comparison between likelihoods  presumably the smaller region fits better in the common are and worse outside. Recommendation: leave as is pending further investigation. One common feature is that there is a nonnegligible fraction of objects (usually with faint mag_psf) with very small differences in initial fit likelihood but large ones in the dev/exp fit, which should not occur frequently. 
I ran multiBandDriver.py on HSCR 9813 3,4 to determine if any cModel configuration parameters can/should be changed to improve the running time and/or bestfit model parameters. My conclusions for the tested parameters are:
initial.nComponents (default 3): For 2, initial fits run in 0.67x the time, but the exp/dev fits run more slowly, so the total fit time is 0.94x. Initial fits are slightly worse on average (with large scatter) and a sizeable fraction of the dev/exp fits are significantly worse (deltalnP < 10). For nComponents=6, the initial fitting time is more than twice as long, with modest increases in the exp/dev lnP (<1 for most galaxies, with scatter). Recommendation: leave it at 3 or lower to 2. initial.optimizer.gradientThreshold (default 1e2): 1e3 runs very slightly faster than default and very slightly improves fits. 1e1 is slightly slower and also results in slightly worse initial/final fits. Recommendation: lower to 1e3. initial.optimizer.maxInnerIterations (default 20): Zero impact on fits, running time for 8 is virtually identical; 50 is 1% slower. Recommendation: leave at 20. initial.optimizer.noSR1Term (default False): True radically changes fitting times by factors of (0.33, 4.37, 4.98)x for (initial, exp, dev) i.e. the initial fit is faster but the subsequent exp/dev are much slower. Confusingly, the initial fits are better on average, whereas the final dev/exp are worse  this may warrant followup. Many of the very slow fits have faint PSF mags and may be pathological cases. Recommendation: leave as False. region.nFitRadiiMax (default 3): Surprisingly, nfitrad=2 initial fits are 0.5% slower than the default, whereas nfitrad=5 is 2% faster. However, the total fitting time scales by a factor of (0.86, 1.38) for the smaller/larger size. Smaller fitting regions naturally have larger likelihoods, but I can't make a meaningful comparison between likelihoods  presumably the smaller region fits better in the common are and worse outside. Recommendation: leave as is pending further investigation. One common feature is that there is a nonnegligible fraction of objects (usually with faint mag_psf) with very small differences in initial fit likelihood but large ones in the dev/exp fit, which should not occur frequently. The quoted figures can be generated by the notebook created in 
Reviewers  Yusra AlSayyad [ yusra ]  Jim Bosch [ jbosch ] 
Status  In Progress [ 3 ]  In Review [ 10004 ] 
Story Points  4  7 
Sprint  DRP S185, DRP S186 [ 707, 727 ]  DRP S185, DRP S186, DRP F181 [ 707, 727, 742 ] 
Status  In Review [ 10004 ]  Reviewed [ 10101 ] 
Resolution  Done [ 10000 ]  
Status  Reviewed [ 10101 ]  Done [ 10002 ] 