compareRootHistos.py

Go to the documentation of this file.

#! /usr/bin/env python
from optparse import OptionParser
import re
import sys
# Well known trick
backupArgv = sys.argv[:]
sys.argv = []
from ROOT import TFile
sys.argv = backupArgv

# The list of histograms not to check. Expressed as regexps
gRegexBlackList = []
histos = ['TH1D', 'TH1F', 'TH2D', 'TH2F', 'TProfile']
ref = 'REFERENCE'
test = 'TEST'

# =============================================================================
# Method   : rec( o, path=None, lst=None )
#
# @param o    : a ROOT object
# @param path : a string like a transient store path; ie '/stat/CaloPIDs/ECALPIDE'
# @param lst  : a list to hold (path, object) tuples
#
# function    : recursively pull apart a ROOT file, making a list of (path, TObject) tuples
#               This is done by GetListOfKeys method, which lets one work down through directories
#               until you hit the Histo at the end of the path.  The list of tuples is returned
#


def rec(o, path=None, lst=None):
    if not path:
        path = '/stat'
        lst = []
    else:
        path = path + '/' + o.GetName()
    lst.append((path, o))
    if 'GetListOfKeys' in dir(o):
        keys = o.GetListOfKeys()
        for k in keys:
            name = k.GetName()
            rec(o.Get(name), path, lst)
    else:
        pass
    return lst
# =============================================================================

# =============================================================================
# Method   : composition( t )
#
# @param t : a tuple of ( type, d ) where type is either 'REFERENCE' or 'TEST'
#            and d is a dictionary of ROOT objects, with each key = ROOT path
#
# function : deduce the composition, (objects/histos) counts
#


def composition(t):
    typ, d = t
    hists = 0
    objs = 0
    for k in d.keys():
        if d[k].__class__.__name__ in histos:
            hists += 1
        else:
            objs += 1
    return objs, hists
# =============================================================================

# =============================================================================
# Method        : comparePaths( t1, t2 )
#
# @param t1, t2 : a tuple of ( type, d ) where type is either 'REFERENCE' or 'TEST'
#                 and d is a dictionary of ROOT objects, with each key = ROOT path
#
# function      : compare the paths between the two histo files.  If the files are identical, they
#                 should have the same set of paths.  The Test file should definitely have the
#                 same paths as the Reference.  Perhaps the Reference file will have some more paths due
#                 to extra histos added as part of Application Sequencer finalisation
#                 Arguments t1 and t2 are checked and the test/reference auto-detected
#


def comparePaths(t1, t2):
    if t1[0] == ref:
        ds = t1[1]
        dp = t2[1]
    elif t2[0] == ref:
        ds = t2[1]
        dp = t1[1]
    else:
        print 'Neither tuple is Reference Root file reference?'
        return

    dsks = ds.keys()
    dpks = dp.keys()
    dsks.sort()
    dpks.sort()

    sset = set(dsks)
    pset = set(dpks)
    os, hs = composition((ref, ds))
    op, hp = composition((test, dp))
    print '\n' + '=' * 80
    print 'Comparison of Paths : Reference vs Test ROOT files'
    print '-' * 80
    print 'Number of paths in Reference file : %i (objects, histos) = ( %i, %i )' % (
        len(dsks), os, hs)
    print 'Number of paths in Test file : %i (objects, histos) = ( %i, %i )' % (
        len(dpks), op, hp)
    matching = sset.intersection(pset)
    matchingHistos = 0
    for n in matching:
        if ds[n].__class__.__name__ in histos:
            matchingHistos += 1
    print '\nMatching paths                 : %i' % (len(matching))
    uSer = sset - pset
    # work out histos unique to test file
    uniqueReferenceHistos = 0
    for n in uSer:
        if ds[n].__class__.__name__ in histos:
            uniqueReferenceHistos += 1
    print 'Paths unique to Reference file : %i ( %i Histos )' % (
        len(uSer), uniqueReferenceHistos)
    if uSer:
        for n in uSer:
            print '\t%s : \t%s' % (ds[n], n)
    uPar = pset - sset
    uniqueTestHistos = 0
    for n in uPar:
        if dp[n].__class__.__name__ in histos:
            uniqueTestHistos += 1
    print 'Paths unique to Test file : %i ( %i Histos )' % (
        len(uPar), uniqueTestHistos)
    if uPar:
        for n in uPar:
            print '\t%s : \t%s' % (dp[n], n)
    print 'Matching Histos to test : %i' % (matchingHistos)
    print '=' * 80 + '\n'
    return (((os, hs), (op, hp)), (uSer, uniqueReferenceHistos), (uPar, uniqueTestHistos), matchingHistos)
# =============================================================================

# =============================================================================
# Method        : bin2binIdentity(h1,h2)
#
# @param h1, h2 : The two histogtams to compare
# function      : Return the number of different bins


def bin2binIdentity(h1, h2):
    def getNbins(h):
        biny = h.GetNbinsY()
        if biny > 1:
            biny += 1
        binz = h.GetNbinsZ()
        if binz > 1:
            binz += 1
        return (h.GetNbinsX() + 1) * (biny) * (binz)

    nbins = getNbins(h1)
    diffbins = 0
    for ibin in xrange(0, nbins):
        h1bin = h1.GetBinContent(ibin)
        h2bin = h2.GetBinContent(ibin)
        diffbins += (h1bin != h2bin)
    return diffbins


# =============================================================================
# Method        : compareHistos( t1, t2 )
#
# @param t1, t2 : a tuple of ( type, d ) where type is either 'REFERENCE' or 'TEST'
#                 and d is a dictionary of ROOT objects, with each key = ROOT path
#
# function      : compare the histograms in Reference/Test ROOT files.  First, go through each
#                 dict to collect the histos (ignore TDirectory objects, etc).  Then the histos
#                 in the test file (experimental) are compared to their equivalents in the
#                 reference file (definitely correct) using 3 methods.
#                 1) The entries are checked, they should be equal
#                 2) If entries are equal, check the Integral(); should be equal
#                 3) If integrals are equal, check the KolmogorovTest() ; should be 1
#                 4) If identity flag is there and KS test is performed, perform bin2bin identity test
#                 Arguments t1 and t2 are checked and the test/reference auto-detected
#
def compareHistos(t1, t2, state, checkBin2BinIdentity):

    (((referenceObjects, referenceHistos), (parallObjects, parallHistos)),
     (uniqueSerPaths, uniqueSerHistos), (uniqueParPaths, uniqueParHistos), mh) = state

    # deduce which one is test, which reference
    if t1[0] == ref:
        ds = t1[1]
        dp = t2[1]
    elif t2[0] == ref:
        ds = t2[1]
        dp = t1[1]
    else:
        print 'Neither tuple is Reference Root file reference?'
        return

    # histocount, objectcount for test/reference
    hcp = 0
    pHistos = []
    hcs = 0
    sHistos = []

    omit = [re.compile(regex) for regex in gRegexBlackList]

    # find the histos in the reference file
    for k in ds.keys():
        if not any(regex.search(k) != None for regex in omit):
            if ds[k].__class__.__name__ in histos:
                hcs += 1
                sHistos.append(k)
    # same for test
    for k in dp.keys():
        if not any(regex.search(k) != None for regex in omit):
            if dp[k].__class__.__name__ in histos:
                hcp += 1
                pHistos.append(k)

    cEntries = 0
    xEntries = 0
    diffEntries = []
    cIntegrals = 0
    xIntegrals = 0
    diffIntegrals = []
    passedKol = 0
    failedKol = 0
    diffKols = []
    zeroIntegrals = 0
    passedIdentity = 0
    failedIdentity = 0
    diffIdentity = []
    identityDiffBins = {}
    kTested = 0
    kTestResults = {}
    notfound = 0
    integralMatch = 0
    otherTest = 0
    zeroIntegralMatch = 0
    for h in sHistos:
        if h in pHistos:
            # matching histos to check
            cEntries += 1
            sh = ds[h]
            ph = dp[h]
            # first check entries
            if sh.GetEntries() != ph.GetEntries():
                diffEntries.append(h)
                xEntries += 1
                continue
            # check for (non-zero sum of bin error) && (non-zero integrals) for K-Test
            sBinError = 0.0
            pBinError = 0.0
            for i in xrange(sh.GetNbinsX()):
                sBinError += sh.GetBinError(i)
            for i in xrange(ph.GetNbinsX()):
                pBinError += ph.GetBinError(i)
            sint = sh.Integral()
            pint = ph.Integral()
            doKS = (bool(sint) and bool(pint)) and (
                sBinError > 0 and pBinError > 0)
            if checkBin2BinIdentity and doKS:
                diffBins = bin2binIdentity(sh, ph)
                if diffBins == 0:
                    passedIdentity += 1
                else:
                    failedIdentity += 1
                    diffIdentity.append(h)
                    identityDiffBins[h] = diffBins
            if (bool(sint) and bool(pint)) and (sBinError > 0 and pBinError > 0):
                kTested += 1
                kTest = sh.KolmogorovTest(ph)
                kTestResults[h] = kTest
                if int(kTest):
                    passedKol += 1
                else:
                    # ; print 'KTest result : ', kTest
                    failedKol += 1
                    diffKols.append(h)
            else:
                # try the integral test?
                otherTest += 1
                if all((sint, pint)) and (sint == pint):
                    integralMatch += 1
                elif (sint == pint):
                    zeroIntegralMatch += 1
                else:
                    diffIntegrals.append(h)
                    xIntegrals += 1
        else:
            notfound += 1
            print 'not found? ', h

    # report on Failed Entry-Checks
    print '\n\n' + '-' * 80
    print 'Summary of histos with different Entries'
    print '-' * 80
    if diffEntries:
        diffEntries.sort()
        for e in diffEntries:
            print '\t\t\t%s:\t%i != %i' % (
                e, int(ds[e].GetEntries()), int(dp[e].GetEntries()))
    print '-' * 80

    # report on Failed Kolmogorov Tests
    print '\n\n' + '-' * 60
    print 'Summary of histos which failed Kolmogorov Test'
    print '-' * 60
    if diffKols:
        diffKols.sort()
        for e in diffKols:
            result = kTestResults[e]  # DP Calculated twice ARGH!!
            print '%s\t\t%s :\tK-Test Result :\t %5.16f' % (
                ds[e].ClassName(), e, result)
    print '-' * 60

    # report on Failed Integral Checks
    print '\n\n' + '-' * 60
    print 'Summary of histos which failed Integral Check'
    print '-' * 60
    if diffIntegrals:
        diffIntegrals.sort()
        for e in diffIntegrals:
            diff = dp[e].Integral() - ds[e].Integral()
            pc = (diff * 100) / ds[e].Integral()
            print '%s\t\t%s:\t Diff = %5.6f\tPercent Diff to Reference : %5.6f ' % (
                ds[e].ClassName(), e, diff, pc)
    print '-' * 60 + '\n'
    print '=' * 80 + '\n'

    # Report on failed bin2bin identity
    if checkBin2BinIdentity:
        # report on b2b checks
        print '\n\n' + '-' * 80
        print 'Summary of histos with at least one bin with different Entries'
        print '-' * 80
        if diffIdentity:
            diffIdentity.sort()
            for e in diffIdentity:
                print '%s\t\t%s: %i different bins' % (
                    ds[e].ClassName(), e, identityDiffBins[e])
            print '-' * 80

    print '\n' + '=' * 80
    print 'Comparison : Reference/Test ROOT Histo files'
    print '\n\t\tReference\tTest'
    print '\tObjects : %i\t%i\t\t( p-s = %i )' % (referenceObjects,
                                                  parallObjects, parallObjects - referenceObjects)
    print '\tHistos  : %i\t%i\t\t( p-s = %i )' % (referenceHistos,
                                                  parallHistos,  parallHistos - referenceHistos)
    print '\t          __________'
    print '\tTotal   : %i\t%i\n' % (
        referenceHistos + referenceObjects,  parallHistos + parallObjects)
    print 'Objects/Histos unique to Reference File : %i / %i' % (
        len(uniqueSerPaths) - uniqueSerHistos, uniqueSerHistos)
    print 'Objects/Histos unique to Test File : %i / %i' % (
        len(uniqueParPaths) - uniqueParHistos, uniqueParHistos)
    print '\nMatching Histograms valid for Comparison : %i' % (mh)
    print '\nOmissions\' patterns : '
    for entry in gRegexBlackList:
        print '\t%s' % (entry)
    print '\nHistograms for Comparison (after Omissions) : %i' % (
        mh - len(gRegexBlackList))
    print '\n\tHISTOGRAM TESTS : '
    print '\t\tKOLMOGOROV TEST      : %i' % (kTested)
    print '\t\tINTEGRAL TEST        : %i' % (otherTest)
    print '\t\tENTRIES TEST         : %i' % (xEntries)
    if checkBin2BinIdentity:
        print '\t\tBIN2BIN TEST         : %i' % (passedIdentity)
    print '\t\t                       ____'
    print '\t\tTested               : %i' % (cEntries)

    print '\n\tDISCREPANCIES : '
    print '\t\tK-Test      : %i' % (failedKol)
    print '\t\tIntegrals   : %i' % (xIntegrals)
    print '\t\tEntries     : %i' % (xEntries)
    retval = failedKol + xIntegrals + xEntries + failedIdentity
    if retval != 0:
        print '\nThe two sets of histograms were not identical'
    print '\n' + '=' * 80
    return retval

# =============================================================================


def extractBlacklist(listString):
    global gRegexBlackList
    if listString:
        for blackRegexp in listString.split(","):
            gRegexBlackList.append(blackRegexp)
    else:
        gBlackList = []

# =============================================================================


if __name__ == '__main__':
    usage = "usage: %prog testFile.root referenceFile.root [options]"
    parser = OptionParser()
    parser.add_option("-b", dest="blacklist",
                      help='Comma separated list of regexps matching histograms to skip (for example -b"MemoryTool,ProcTime")')

    parser.add_option("-i",
                      action="store_true", dest="bin2bin", default=False,
                      help="Check for bin to bin identity")
    (options, args) = parser.parse_args()

    if len(args) != 2:
        print "Wrong number of rootfiles. Usage:"
        print usage
        sys.exit(1)

    extractBlacklist(options.blacklist)

    testFile, referenceFile = args

    tfs = TFile(testFile, 'REC')
    print 'opening Test File : %s' % (testFile)
    tfp = TFile(referenceFile, 'REC')
    print 'opening Reference File : %s' % (referenceFile)

    # get structure of TFiles in a list of (path, object) tuples
    lref = rec(tfs)
    ltest = rec(tfp)
    # make a dictionary of lser and lpar.  keys=paths
    dref = dict([(n, o) for n, o in lref])
    dtest = dict([(n, o) for n, o in ltest])
    # make a tuple of (type, dict) where type is either 'reference' or 'test'
    ts = (ref, dref)
    tp = (test, dtest)

    # check objs/histos in each file
    composition(ts)
    composition(tp)

    # compare paths from each file
    state = comparePaths(ts, tp)

    # compare histos from each file
    retval = compareHistos(ts, tp, state, checkBin2BinIdentity=options.bin2bin)

    # finished with TFiles
    tfs.Close()
    tfp.Close()

    sys.exit(retval)