BPZ v1.99.3 updates

Major bug fix from v1.99.2 (both v1.98b and my UDF analysis Coe06 are unaffected):
Non-detections were handled incorrectly (as non-observations instead) in bpz_tools.py:

       #nonobs=less(reshape(f, (1, 1, self.nf)) / reshape(ef, (1, 1, self.nf)) + ft_z*0., 1e-4)
# Above was wrong: non-detections were ignored as non-observed --DC
nonobs=greater(reshape(ef, (1, 1, self.nf)) + ft_z*0., 1.0)

Bug fix: Small bias of 0.01 in redshift corrected.

Also new in v1.99.3:

Photometric calibration tests -- improved reporting (and now default to do so)
-- tests whether individual filters may be miscalibrated relative to the rest

Improved plotting:
- SED fits plotted in AB magnitude rather than [erg / cm^2 / s / Angstrom]
- webpage to summarize results: SED fits and probablility plots

Priors now more flexible, allowing for variable numbers of elliptical, spiral, and starburst/irregular templates.  (Set these numbers using -NTYPES.  If not set, assumes 1 elliptical SED, 2 spiral, and the rest starburst/irregular.)

-PROBS2 output (experimental): gives full P(z,t) but much more compact than -PROBS


A few small changes I made since the original v1.99.3 release (4/30/10):
(all included in the current v1.99.3 release)

5/9/10
SED/Vega_reference.sed & flat.sed re-included in distribution (were mistakenly left out)

5/7/10
sedplotAB.py:
 - non-detections & non-observations now plotted
 - better fitting of gray line to blue rectangles
 - log x-axis was still default in some ways.  fixed that

coeplot.pyploterrorbars: removed a print statement (dxlog)
coeio.py: loadpixelscale fix (for ColorPro)

5/6/10
webpage.py
: added ability to plot first N objects

Earlier changes from 1.98b -> 1.99.2:

Python package dependencies:
v1.98b uses Python's Numeric & biggles
v1.99.2 uses Python's Numpy & matplotlib

N_PEAKS: summarize information for multiple peaks in P(z)

By default, all P(z) are convolved with a Gaussian of width 0.03:  -CONVOLVE_P yes

Due to numerical instabilities (Coe06), the output is "not quite Bayesian" (though the prior is still used):  bpz.py calls bpz_tools.py's likelihood.likelihood instead of likelihood.Bayes_likelihood

In both versions, the range zbmin--zbmax is artificially inflated (by default).  We have decreased the default 1-sigma spread from 0.067 to 0.05:  -MIN_RMS 0.05

Changed default inputs:
-SPECTRA CWWSB.list -> CWWSB4.list
-PRIOR hdfn -> hdfn_SB -> hdfn_gen
-ZMAX 6.5 -> 10
-MIN_RMS 0.067 -> 0.05

New options added:
    'MIN_MAGERR':   0.001
    'PHOTO_ERRORS':'no',    #Define the confidence interval using only the photometric errors
    'N_PEAKS':1,
    'MERGE_PEAKS':'no',
    'CONVOLVE_P':'yes',
    'P_MIN':1e-2,
    'SED_DIR': sed_dir,
    'AB_DIR': ab_dir,
    'FILTER_DIR': fil_dir,
    'DELTA_M_0': 0.,
    'ZP_OFFSETS':0.,
    'ZC': None,
    'FC':None,
    "ADD_SPEC_PROB":None,
    "ADD_CONTINUOUS_PROB":None,
    "NMAX": None # Useful for testing