Likelihood Test Results (3)

As a sanity check I re-computed the likelihoods with the old qso catalog to make sure I got the same results as before, which I did. See ../logs/100426log.pro for code.

It seems that adding in the BOSS QSOs doesn't help.

Now I am going to run on the old qso catalog but with the larger redshift range (0.5 < z < 5.0) and see if this improves things. See ../logs/100426log.pro for code.

So this seems to actually make a difference. I have played around with the redshift range for the numerator to get the best selection numbers:

eps = 1e-30
bossqsolike = total(likelihood.L_QSO_Z[14:34],1) ;quasars between redshift 2.1 and 3.5
qsolcut1 = where( alog10(total(likelihood.L_QSO_Z[19:34],1)) LT -9.0)
den = total(targets.L_EVERYTHING_ARRAY[0:4],1) + total(likelihood.L_QSO_Z[0:44],1) + eps
num = bossqsolike + eps
NEWRATIO = num/den
NEWRATIO[qsolcut1] = 0 ; eliminate objects with low L_QSO value


qsolcut2 = where( alog10(total(targets.L_QSO_Z[0:18],1)) LT -9.0)
L_QSO = total(targets.L_QSO_Z[2:18],1)
den = total(targets.L_EVERYTHING_ARRAY[0:4],1) + total(targets.L_QSO_Z[0:18],1) + eps
num = L_QSO + eps
OLDRATIO = num/den
OLDRATIO[qsolcut2] = 0 ; eliminate objects with low L_QSO value


numsel = 1800-1
NRend = n_elements(newratio)-1
sortNR = reverse(sort(newratio))
targetNR = sortNR[0:numsel]
restNR = sortNR[numsel+1:NRend]

ORend = n_elements(oldratio)-1
sortOR = reverse(sort(oldratio))
targetOR = sortOR[0:numsel]
restOR = sortOR[numsel+1:ORend]

nql = setintersection(quasarindex,targetNR)
nqnl = setintersection(quasarindex, restNR)
nsl = setdifference(targetNR, quasarindex)
nsnl = setdifference(restNR, quasarindex)

ql = setintersection(quasarindex, targetOR)
qnl = setintersection(quasarindex, restOR)
sl = setdifference(targetOR,quasarindex)
snl = setdifference(restOR,quasarindex)

IDL> print, n_elements(ql) ; number quasars
461
IDL> print, n_elements(ql)*1.0/(n_elements(sl)+n_elements(ql)) ;percent accuracy
0.256111
IDL> print, n_elements(ql) + n_elements(sl) ; total targeted
1800
IDL>
IDL> print, n_elements(nql) ; number quasars
508
IDL> print, n_elements(nql)*1.0/(n_elements(nsl)+n_elements(nql)) ;percent accuracy
0.282222
IDL> print, n_elements(nql) + n_elements(nsl) ; total targeted
1800

So we go from 25.6% accuracy to 28.2% accuracy! And we find 47 more quasars!

Plots
The white points were targeted/missed by both the new and old likelihoods
The magenta points were only targeted/missed by the new likelihood
The cyan points were only targeted/missed by the old likelihood

The code for the above run is in the directory ../likelihood/run1/

Now to see how the different luminosity functions do. Below is results running with the Richard 06 luminosity function:

QSO Catalog with Richards Luminosity Function

(See ../logs/100427_2log.pro for code)

IDL> print, n_elements(ql) ; number quasars
461
IDL> print, n_elements(ql)*1.0/(n_elements(sl)+n_elements(ql)) ;percent accuracy
0.256111
IDL> print, n_elements(ql) + n_elements(sl) ; total targeted
1800
IDL>
IDL> print, n_elements(nql) ; number quasars
543
IDL> print, n_elements(nql)*1.0/(n_elements(nsl)+n_elements(nql)) ;percent accuracy
0.301667
IDL> print, n_elements(nql) + n_elements(nsl) ; total targeted
1800

This does better than the old luminosity function!
So we go from 25.6% accuracy to 30.2% accuracy! And we find 82 more quasars!

Plots
The white points were targeted/missed by both the new and old likelihoods
The magenta points were only targeted/missed by the new likelihood
The cyan points were only targeted/missed by the old likelihood