Changed generation PDF in randomized regression

In ANNZ::getRndMethodBestPDF(), the PDF was constructed by selecting
the MLMs with the best performance in terms of scatter and outlier
fractions. This has been changed to take into account the combination
of scatter and bias instead.

Also added the user-parameters: “max_sigma68_PDF”, “max_bias_PDF” and
“max_frac68_PDF”. These allow the user to define the maximal value of
the scatter, bias and combined outlier-fraction metrics of MLMs which
are included in PDFs in randomized regression.

examples/scripts/annz_rndReg_advanced.py

@@ -352,6 +352,16 @@
   # -----------------------------------------------------------------------------------------------------------
   glob.annz["addOutputVars"] = "MAG_U;MAGERR_I"
 
+  # -----------------------------------------------------------------------------------------------------------
+  # max_sigma68_PDF, max_bias_PDF, max_frac68_PDF
+  #   - if max_sigma68_PDF,max_bias_PDF are positive, they put thresholds on the maximal value of
+  #     the scatter (max_sigma68_PDF), bias (max_bias_PDF) or outlier-fraction (max_frac68_PDF) of an
+  #     MLM which may be included in the PDF created in randomized regression
+  # -----------------------------------------------------------------------------------------------------------
+  glob.annz["max_sigma68_PDF"] = 0.044
+  glob.annz["max_bias_PDF"]    = 0.01
+  glob.annz["max_frac68_PDF"]  = 0.08
+
 
   # --------------------------------------------------------------------------------------------------
   # optimization

examples/scripts/generalSettings.py

@@ -51,9 +51,6 @@ def generalSettings():
   #   the metric which is more significant in deciding which MLM performs "best".
   # glob.annz["optimCondReg"] = "bias"
 
-  # option to only generate randomized PDFs based on the scatter metric, not considering outlier fractions
-  # glob.annz["doNotUseOutFracPdfWeights"] = True
-
   # number of random MLM weighting schemes to generate as part of getRndMethodBestPDF()
   # glob.annz["nRndPdfWeightTries"] = 50
 

include/ANNZ.hpp

@@ -144,7 +144,7 @@ class ANNZ : public BaseClass {
   void     getBestANNZ(map < int,vector<int> >    & zRegQnt_nANNZ,   map < int,vector<double> > & zRegQnt_bias,
                        map < int,vector<double> > & zRegQnt_sigma68, map < int,vector<double> > & zRegQnt_fracSig68,
                        vector < int >             & bestMLMsV,       bool                       onlyInclusiveBin = true);
-  void     getRndMethodBestPDF(TTree * aChain, int bestANNZindex, vector<int>  & zRegQnt_nANNZ,
+  void     getRndMethodBestPDF(TTree * aChain, int bestANNZindex, vector<int>  & zRegQnt_nANNZ, vector<double> & zRegQnt_bias, 
                                vector<double> & zRegQnt_sigma68, vector<double> & zRegQnt_fracSig68, vector < vector<double> > & bestWeightsV);
   void     doEvalReg(TChain * inChain = NULL, TString outDirName = "", vector <TString> * selctVarV = NULL);
   void     doMetricPlots(TChain * inChain = NULL, vector <TString> * selctMLMv = NULL);