Esempio n. 1
0
 def __init__(self):
     self.usage = "usage: BundleHistos [options]"
     #define the parser with options only relevant for the digester
     self.parser = OptionParser(self.usage)
     self.parser.add_option("-f", "--file", dest="filename",action="store",
                            help="ini file(s) to read root object from ")
     self.parser.add_option("-o", "--outputname", dest="outputname",action="store",
                            default="./fitterInputs.root", help="name of root file to dump to")
     self.parser.add_option("-l", "--fileHistoList", dest="fileHistoList",action="store",
                            help="semi colon separated list of triple <file:histoName:newName>")
     self.objectDB         = RootObjectLoader()
     self.configFile = None
     self.fileHistoList    = []
     self.outputname = None
     self.loaded = None
     self.openedFiles = []
     self.nBins = 0
     self.obsName = ""
     self.nFits = 0
     self.fitNames = []
     self.BinBorders = set()
     self.UpperBinLimit = 0
     self.LowerBinLimit = 0
     self.Histos2Write = []
     self.Graphs2Write = []
     self.outFile = None
Esempio n. 2
0
    def __init__(self):
        self.usage = "usage: collectNumbersToTGraph.py [options]"
        #define the parser with options only relevant for the digester
        self.parser = OptionParser(self.usage)
        self.parser.add_option("-f", "--file", dest="filename",action="store",
                               help="ini file(s) to read root object from ")
        self.parser.add_option("-o", "--outputname", dest="outputname",action="store",
                               default="./fitterInputs.root", help="name of root file to dump to")
        self.parser.add_option("-l", "--fileHistoList", dest="fileHistoList",action="store",
                               help="semi colon separated list of triple <file:histoName:newName>")
        self.parser.add_option("-E", "--extractor", dest="extractor",action="store",
                               help="extractor object to use on inline histo for value,error yields, see %s/python/RootObjectExtractors.py for details" % basePath)

        self.objectDB         = RootObjectLoader()
        self.configFile = None
        self.fileHistoList    = []
        self.outputname = None
        self.dictToWrite = None
        self.openedFiles = []
        self.extractor = None
        self.extractorString = "TotalIntegralExtractor()"
Esempio n. 3
0
class BundleHistos:
    
    def __init__(self):
        self.usage = "usage: BundleHistos [options]"
        #define the parser with options only relevant for the digester
        self.parser = OptionParser(self.usage)
        self.parser.add_option("-f", "--file", dest="filename",action="store",
                               help="ini file(s) to read root object from ")
        self.parser.add_option("-o", "--outputname", dest="outputname",action="store",
                               default="./fitterInputs.root", help="name of root file to dump to")
        self.parser.add_option("-l", "--fileHistoList", dest="fileHistoList",action="store",
                               help="semi colon separated list of triple <file:histoName:newName>")
        self.objectDB         = RootObjectLoader()
        self.configFile = None
        self.fileHistoList    = []
        self.outputname = None
        self.dictToWrite = None
        self.openedFiles = []

    def initFromInput(self):
        print "%s >> parseInput" % self.__class__
        popts,pargs = self.parser.parse_args()
        self.parsedOpts = popts.__dict__
        self.parsedArgs = pargs
        if "help" in self.parsedOpts.keys() or "h" in self.parsedOpts.keys() or not self.parsedOpts.keys():
            self.parser.print_help()
            return 1
        
        #do all flags
        if self.parsedOpts['outputname']:
            self.outputname = self.parsedOpts['outputname']

        if self.parsedOpts['filename']:
            self.configFile = self.parsedOpts['filename']
            filename = self.configFile.split("/")[-1]
            print "opening DB from",self.configFile 
            self.objectDB.setConfigFile(self.configFile)
            self.objectDB.openFiles(False)
            self.initDictFromDatabase()
        else:
            if self.parsedOpts['fileHistoList']:
                self.initDictFromList(self.parsedOpts['fileHistoList'])
            else:
                raise Exception("no input given, exiting")


    def initDictFromList(self,_text=""):
        """
        function to load dictToWrite from list given by CLI
        """
        self.dictToWrite = {}
        files = [ item.split(":")[0] for item in _text.split(";") ]
        objects = [ item.split(":")[1] for item in _text.split(";") ]
        names = [ item.split(":")[2] for item in _text.split(";") ]
        self.openedFiles = [ ROOT.TFile.Open(item) for item in files if os.path.isfile(item) ]
        for name in objects:
            index = objects.index(name)
            meta = self.openedFiles[index].Get(name)
            if meta.__nonzero__():
                self.dictToWrite[names[index]] = meta
            else:
                print "could not load %s from %s" % (name,files[index])
        
        return 0

    def initDictFromDatabase(self):
        """
        function to load dictToWrite from objectDB
        """
        self.dictToWrite = {}
        #print self.objectDB.loadedCommands
        for key,value in self.objectDB.loadedDict.iteritems():
            for item in value:
                self.dictToWrite[key] = item

            # if self.objectDB.loadedCommands.has_key(key):
            #     for cmd in self.objectDB.loadedCommands[key]:
            #         try:
            #             eval(cmd.replace("[]","self.dictToWrite["+key+"]"))
            #         except:
            #             print ">> command %s failed on object named [%s], %s" % (cmd,key,self.dictToWrite[key].GetName())
            #         else:
            #             print ">> command %s succeeded for [%s]" % (cmd,key)
            # else:
            #     print ">> no commands found for ",key
        return 0

        
    def Run(self):

        if not self.dictToWrite:
            print "no objects loaded"
            return
            
        nfile = ROOT.TFile.Open(self.outputname,"RECREATE")

        for key,value in self.dictToWrite.iteritems():
            meta = value.Clone(key)
            meta.SetDirectory(nfile)
            print "adding %s to %s" % (value,self.outputname)
            meta.Write()

        nfile.Close()
        self.objectDB.close()
        if self.openedFiles:
            [ item.Close() for item in self.openedFiles ]
Esempio n. 4
0
class BundleHistos:
    
    def __init__(self):
        self.usage = "usage: BundleHistos [options]"
        #define the parser with options only relevant for the digester
        self.parser = OptionParser(self.usage)
        self.parser.add_option("-f", "--file", dest="filename",action="store",
                               help="ini file(s) to read root object from ")
        self.parser.add_option("-o", "--outputname", dest="outputname",action="store",
                               default="./fitterInputs.root", help="name of root file to dump to")
        self.parser.add_option("-l", "--fileHistoList", dest="fileHistoList",action="store",
                               help="semi colon separated list of triple <file:histoName:newName>")
        self.objectDB         = RootObjectLoader()
        self.configFile = None
        self.fileHistoList    = []
        self.outputname = None
        self.loaded = None
        self.openedFiles = []
        self.nBins = 0
        self.obsName = ""
        self.nFits = 0
        self.fitNames = []
        self.BinBorders = set()
        self.UpperBinLimit = 0
        self.LowerBinLimit = 0
        self.Histos2Write = []
        self.Graphs2Write = []
        self.outFile = None

    def initFromInput(self):
        print "%s >> parseInput" % self.__class__
        popts,pargs = self.parser.parse_args()
        self.parsedOpts = popts.__dict__
        self.parsedArgs = pargs
        if "help" in self.parsedOpts.keys() or "h" in self.parsedOpts.keys() or not self.parsedOpts.keys():
            self.parser.print_help()
            return 1
        
        #do all flags
        if self.parsedOpts['outputname']:
            self.outputname = self.parsedOpts['outputname']

        if self.parsedOpts['filename']:
            self.configFile = self.parsedOpts['filename']
            filename = self.configFile.split("/")[-1]

            self.objectDB.setConfigFile(self.configFile)
            #self.objectDB.openFiles(False)
            self.initDictFromDatabase()
        else:
            if self.parsedOpts['fileHistoList']:
                self.initDictFromList(self.parsedOpts['fileHistoList'])
            else:
                raise Exception("no input given, exiting")


    def initDictFromList(self,_text=""):
        """
        function to load loaded from list given by CLI
        """
        self.loaded = {}
        files = [ item.split(":")[0] for item in _text.split(";") ]
        objects = [ item.split(":")[1] for item in _text.split(";") ]
        names = [ item.split(":")[2] for item in _text.split(";") ]
        self.openedFiles = [ ROOT.TFile.Open(item) for item in files if os.path.isfile(item) ]
        for name in objects:
            index = objects.index(name)
            meta = self.openedFiles[index].Get(name)
            if meta.__nonzero__():
                self.loaded[meta] = names[index]
            else:
                print "could not load %s from %s" % (name,files[index])
        
        return 0

    def initDictFromDatabase(self):
        """
        function to load loaded from objectDB
        """
        self.loaded = {}
        self.loadedKeys = {}
        #for key,value in self.objectDB.loadedDict.iteritems():
        #    for item in value:
        #        self.loaded[item] = key

        for sec in self.objectDB.cfgParser.sections():
            self.loadedKeys[sec] = {}
            for k,v in self.objectDB.cfgParser.items(sec):
                if v.lower().count(".root"):
                    fileName = self.replaceEnvironment(v)
                    print "opening ",fileName
                    self.openedFiles.append(ROOT.TFile.Open(fileName))
                    self.loaded[sec] = self.openedFiles[-1]
                else:
                    self.loadedKeys[sec][k] = v

        return 0

    def replaceEnvironment(self,_text):
        value = _text
        if _text.count("__")>1:
            dirItems = _text.split("/")
            envItem = [item for item in dirItems if item.count("__")>1][0]
            envName = envItem[envItem.find("__")+2:envItem.rfind("__")]
            value = _text.replace(envItem,os.getenv(envName))
            
        return value

    def collectData(self):

        self.nBins = len(self.loadedKeys.keys())
        print "merging into %i bins" % self.nBins
        centralValueObjs = [ item.Get("centralValues") for item in self.openedFiles if item.Get("centralValues").__nonzero__() ]
        if len(centralValueObjs):
            self.nFits = centralValueObjs[0].GetNbinsX()
            for item in range(1,self.nFits+1):
                self.fitNames.append(str(centralValueObjs[0].GetXaxis().GetBinLabel(item)))
        print "found %i fits\n%s" % (self.nFits,self.fitNames)
                
        for item,content in self.loadedKeys.iteritems():
            if not self.obsName and content.has_key("observablename"):
                self.obsName = content["observablename"]
                print "observable to process ", self.obsName

            if content.has_key("binid"):
                self.BinBorders.add(float(content["binid"]))
            elif content.has_key("binvalue"):
                binValue = content["binvalue"]
                self.BinBorders.add(float(binValue.split(",")[0]))
                if len(binValue.split(","))>1:
                    self.BinBorders.add(float(content["binvalue"].split(",")[-1]))
            else:
                print ">> %s does not have binID nor binValue, skipping \n%s" % (item,content)
                
        self.UpperBinLimit = max(self.BinBorders)
        self.LowerBinLimit = min(self.BinBorders)
        print "[%.2f,%.2f] %s" % (self.LowerBinLimit,self.UpperBinLimit,self.BinBorders)
        return

    def createResultHistos(self):
        self.outFile = ROOT.TFile(self.outputname,"RECREATE")
        self.outFile.WriteFree()

        for idx in range(self.nFits):
            currentName = self.fitNames[idx].replace("{","").replace("}","")
            self.Histos2Write.append(ROOT.TH1D(currentName+"_symm",";%s;%s" % (self.obsName,currentName),self.nBins,self.LowerBinLimit,self.UpperBinLimit))
            self.Graphs2Write.append(ROOT.TGraphAsymmErrors(self.nBins))
            self.Graphs2Write[-1].GetYaxis().SetName(self.fitNames[idx])
            self.Graphs2Write[-1].GetYaxis().SetName(self.obsName)
            self.Graphs2Write[-1].SetName(currentName+"_asymm")
            
        return

    def fillResultHistos(self):

        for abin in range(self.nBins):
            currentFile = self.openedFiles[abin]
            print ">> running on file",currentFile.GetName()
            for fit in range(self.nFits):

                central = currentFile.Get("centralValues").GetBinContent(fit+1)
                symmUncert = currentFile.Get("SymmErrors").GetBinContent(fit+1)
                UncertUp = currentFile.Get("AsymmErrorsUp").GetBinContent(fit+1)
                UncertDown = currentFile.Get("AsymmErrorsDown").GetBinContent(fit+1)
                binUncert = self.Histos2Write[fit].GetXaxis().GetBinWidth(abin+1)/2
                binCenter = self.Histos2Write[fit].GetXaxis().GetBinCenter(abin+1)

                self.Histos2Write[fit].SetBinContent(abin+1,central)
                self.Histos2Write[fit].SetBinError(abin+1,symmUncert)
                self.Graphs2Write[fit].SetPoint(abin,binCenter,central)
                self.Graphs2Write[fit].SetPointError(abin,binUncert,binUncert,UncertDown,UncertUp)
                print "[f:%i,bin:%i] %s: added [x,y] = %.2f, %.2f; sigma(x) = %.2f, sigma(y) = %.2f, <up/down> = %.2f/%.2f" % (fit,abin,self.fitNames[fit], binCenter,central,binUncert,symmUncert,UncertUp,UncertDown)

        return
    
    def writeAll(self):
        
        for graph in self.Graphs2Write:
            graph.Write()
        self.outFile.Write()
        self.outFile.Close()

        return
    

    def Run(self):

        if not self.loaded:
            print "no objects loaded"
            return
            

        try:
            self.collectData()
        except Exception as inst:
            print "collectData failed due to ",inst

        try:
            self.createResultHistos()
        except Exception as inst:
            print "createResultHistos failed due to ",inst
            
        try:
            self.fillResultHistos()
        except Exception as inst:
            print "fillResultHistos failed due to ",inst

        try:
            self.writeAll()
        except Exception as inst:
            print "writeAll failed due to ",inst
Esempio n. 5
0
class BundleHistos:
    
    def __init__(self):
        self.usage = "usage: collectNumbersToTGraph.py [options]"
        #define the parser with options only relevant for the digester
        self.parser = OptionParser(self.usage)
        self.parser.add_option("-f", "--file", dest="filename",action="store",
                               help="ini file(s) to read root object from ")
        self.parser.add_option("-o", "--outputname", dest="outputname",action="store",
                               default="./fitterInputs.root", help="name of root file to dump to")
        self.parser.add_option("-l", "--fileHistoList", dest="fileHistoList",action="store",
                               help="semi colon separated list of triple <file:histoName:newName>")
        self.parser.add_option("-E", "--extractor", dest="extractor",action="store",
                               help="extractor object to use on inline histo for value,error yields, see %s/python/RootObjectExtractors.py for details" % basePath)

        self.objectDB         = RootObjectLoader()
        self.configFile = None
        self.fileHistoList    = []
        self.outputname = None
        self.dictToWrite = None
        self.openedFiles = []
        self.extractor = None
        self.extractorString = "TotalIntegralExtractor()"

    def initFromInput(self):
        print "%s >> parseInput" % self.__class__
        popts,pargs = self.parser.parse_args()
        self.parsedOpts = popts.__dict__
        self.parsedArgs = pargs
        if "help" in self.parsedOpts.keys() or "h" in self.parsedOpts.keys() or not self.parsedOpts.keys():
            self.parser.print_help()
            return 1
        
        #do all flags
        if self.parsedOpts['outputname']:
            self.outputname = self.parsedOpts['outputname']

        if self.parsedOpts['extractor']:
            self.extractorString = self.parsedOpts['extractor']
            

        if self.parsedOpts['filename']:
            self.configFile = self.parsedOpts['filename']
            filename = self.configFile.split("/")[-1]

            self.objectDB.setConfigFile(self.configFile)
            self.objectDB.openFiles(False)
            self.initDictFromDatabase()
        else:
            if self.parsedOpts['fileHistoList']:
                self.initDictFromList(self.parsedOpts['fileHistoList'])
            else:
                raise Exception("no input given, exiting")


    def initDictFromList(self,_text=""):
        """
        function to load dictToWrite from list given by CLI
        """
        self.dictToWrite = {}
        files = [ item.split(":")[0] for item in _text.split(";") ]
        objects = [ item.split(":")[1] for item in _text.split(";") ]
        names = [ item.split(":")[2] for item in _text.split(";") ]
        self.extractor = createExtractor(self.extractorString)
        self.openedFiles = [ ROOT.TFile.Open(item) for item in files if os.path.isfile(item) ]
        for name in objects:
            index = objects.index(name)
            meta = self.openedFiles[index].Get(name)
            if meta.__nonzero__():
                self.dictToWrite[names[index]] = meta
            else:
                print "could not load %s from %s" % (name,files[index])
        
        return 0

    def initDictFromDatabase(self):
        """
        function to load dictToWrite from objectDB
        """
        self.dictToWrite = {}
        #print self.objectDB.loadedCommands
        for key,value in self.objectDB.loadedDict.iteritems():
            for item in value:
                self.dictToWrite[key] = item

            # if self.objectDB.loadedCommands.has_key(key):
            #     for cmd in self.objectDB.loadedCommands[key]:
            #         try:
            #             eval(cmd.replace("[]","self.dictToWrite["+key+"]"))
            #         except:
            #             print ">> command %s failed on object named [%s], %s" % (cmd,key,self.dictToWrite[key].GetName())
            #         else:
            #             print ">> command %s succeeded for [%s]" % (cmd,key)
            # else:
            #     print ">> no commands found for ",key
        return 0

    def getXValueForKey(self,_key):
        
        section = self.objectDB.cfgParser.items(_key)
        
        if not section:
            print "command DB does not contain key ", _key
            print self.objectDB.cfgParser
            return None

        xvalKeys = [ v for (k,v) in section if k.lower().count("x") and k.lower().count("value") ]

        if not xvalKeys:
            print "no xvalue keys found in command section for ", _key
            print section
            return None
        else:
            return float(xvalKeys[0])
        
        
    def Run(self):

        if not self.dictToWrite:
            print "no objects loaded"
            return
        
            
        nfile = ROOT.TFile.Open(self.outputname,"RECREATE")
        tgre = ROOT.TGraphErrors(len(self.dictToWrite.keys()))

        points = []


        for key,value in self.dictToWrite.iteritems():
            print "procesing ",key

            xvalue = self.getXValueForKey(key)
            if not xvalue:
                print "no xvalue found!"
                continue

            section = [ v for (k,v) in self.objectDB.cfgParser.items(key) if v.lower().count("extractor") ]
            if section:
                self.extractorString = section[0]
                self.extractor = createExtractor(self.extractorString)

            self.extractor.setup(value)
            central,error = self.extractor.value(),self.extractor.error()
            print ">> x = %f, y = (%f +/- %f)" % (xvalue,central,error)
            points.append((xvalue,central,error))

        
        for p in sorted(points, key=lambda x: x[0],reverse=False):
            tgre.SetPoint(points.index(p),p[0],p[1])
            tgre.SetPointError(points.index(p),0.,p[-1])

        #draw everything
        acanvas = ROOT.TCanvas(self.outputname.split("/")[-1].replace(".root",""),"",800,600)
        acanvas.Clear()
        acanvas.Draw()
        
        tgre.Draw("AP")
        
        acanvas.Update()
        acanvas.Print(".eps")
        acanvas.SaveAs(self.outputname.replace(".root",".C"))
        tgre.SaveAs(self.outputname.replace(".root","_graph.root"))
        
        
        nfile.Close()
        self.objectDB.close()
        if self.openedFiles:
            [ item.Close() for item in self.openedFiles ]