Friday, August 27, 2010

Recomputing Efficiency and Completeness

I recomputed the Efficiency and Completeness using Stripe 82 BOSS Data:

To determine performance of the likelihood method, we calculated probabilities on single-epoch data in Stripe-82 and compared that target list with the commissioning truth table (which includes targets from all targeting methods and all previously known quasars). Likelihood probabilities were calculated on 112,809 objects and of those the top 1,450 likelihoods were selected for a target density of 40 objects per square-degree. We found an efficiency (# QSOs Found /# Total Targets) of 63% and the completeness (# QSOs Found by likelihood with 40 targets per square-degree / #QSOs found in region targeted by three methods with 80 targets per square-degree) is 55%.

The code to perform this calculation is here:

It is a little strange because the truth table does not include all of stripe 82, so to understand the efficiencies we need to figure out the fraction of Stripe 82 which is covered by the truth table:

fraction = n_elements(targetNRA)*1.0/n_elements(where(stripe82targets.l_ratio GE threshold))*1.0 ;Fraction of stripe 82 observed

fraction = 14.5%

So the total number of targets is not the full number of objects in Erin's median seeing file, but ~14.5% of those objects.

Here are some plots:
Likelihood Selected QSOs:
Likelihood Missed QSOs:
Likelihood Selected/Missed QSOs together:
Likelihood Selected Stars:

All Three:

Code for these plots is also in the following log:

No comments:

Post a Comment