def main(self, conceptMap, symbolMap, symbolDt, dcpts, dpmfs, symbolName, outDir):
conceptMap = LexMap().read(conceptMap)
symbolMap = LexMap().read(symbolMap)
_sink_ = int(symbolMap["_sink_"])
_SINK_ = int(conceptMap["_SINK_"])
conceptCard = len(conceptMap)
symbolCard = len(symbolMap)
jointDtC1C2C3C4 = gmtk.readDt(symbolDt)
jointDcptC1C2C3C4 = gmtk.readDcpt(dcpts, "jointProbC1C2C3C4")
jointProbC1C2C3C4 = gmtk.combineDtDcpt1(jointDtC1C2C3C4, jointDcptC1C2C3C4)
symbolDpmfC1C2C3C4 = gmtk.readDpmf(dpmfs, "%sGivenC1C2C3C4" % symbolName)
symbolGivenC1C2C3C4 = gmtk.combineDtDcpt2(jointDtC1C2C3C4, symbolDpmfC1C2C3C4)
jointProbSymbolC1C2C3C4 = symbolGivenC1C2C3C4.multiple([1, 2, 3, 4], jointProbC1C2C3C4)
gmtk.saveDpmfsProbs(
outDir,
"%sGivenC1C2C3C4" % symbolName,
len(symbolGivenC1C2C3C4.vectSubList([1, 2, 3, 4])),
symbolCard,
symbolGivenC1C2C3C4,
)
jointProbSymbolC1C2C3 = jointProbSymbolC1C2C3C4.marginalize([0, 1, 2, 3])
symbolGivenC1C2C3 = jointProbSymbolC1C2C3.conditionalize([1, 2, 3])
gmtk.saveDpmfsProbs(
outDir,
"%sGivenC1C2C3" % symbolName,
len(symbolGivenC1C2C3.vectSubList([1, 2, 3])),
symbolCard,
symbolGivenC1C2C3,
)
jointProbSymbolC1C2 = jointProbSymbolC1C2C3.marginalize([0, 1, 2])
symbolGivenC1C2 = jointProbSymbolC1C2.conditionalize([1, 2])
gmtk.saveDpmfsProbs(
outDir, "%sGivenC1C2" % symbolName, len(symbolGivenC1C2.vectSubList([1, 2])), symbolCard, symbolGivenC1C2
)
jointProbSymbolC1 = jointProbSymbolC1C2.marginalize([0, 1])
symbolGivenC1 = jointProbSymbolC1.conditionalize([1])
# in case of conditioning by _SINK_ I have to enable to generate _sink_ word only
# otherwise I would see _SINK_ concept in the stack
# TODO: Turn it into validator
symbolGivenC1.setValue([_sink_, _SINK_], 1)
gmtk.saveDcptBigram(outDir, "%sGivenC1" % symbolName, symbolCard, conceptCard, symbolGivenC1)
symbolUnigram = jointProbSymbolC1.marginalize([0])
# I need to enable to decode _unseen_ word only ! so set the probability of
# generating _empty_ to zero
symbolUnigram.setValue([int(symbolMap["_empty_"])], 0)
# normalize sum of probabilities to one
symbolUnigram = symbolUnigram.normJoint()
gmtk.saveDcptUnigram(outDir, "%sUnigram" % symbolName, symbolCard, symbolUnigram)
gmtk.saveDcptUnseen(outDir, "%sZerogram" % symbolName, symbolCard, symbolMap)
if verbose:
print("Start")
print("-------------------------------------------------")
print("Smoothing concept1GivenC2C3C4.")
conceptMap = LexMap().read(conceptFileName)
wordMap = LexMap().read(wordFileName)
conceptCard = len(conceptMap)
wordCard = len(wordMap)
jointDtC1C2C3C4 = gmtk.readDt(wordDtFileName)
jointDcptC1C2C3C4 = gmtk.readDcpt(dcptsFileName, "jointProbC1C2C3C4")
assert toolkit.testProb(jointDcptC1C2C3C4), "Sum of probabilities should be always 1."
jointProbC1C2C3C4 = gmtk.combineDtDcpt1(jointDtC1C2C3C4, jointDcptC1C2C3C4)
# desable generating _EMPTY_ by _EMPTY_, _EMPTY_, _EMPTY_
# jointProbC1C2C3C4.setValue([int(conceptMap["_EMPTY_"]),
# int(conceptMap["_EMPTY_"]),
# int(conceptMap["_EMPTY_"]),
# int(conceptMap["_EMPTY_"])], 0)
##concept1DtC2C3C4 = gmtk.readDt(concept1DtFileName)
##concept1DpmfC2C3C4 = gmtk.readDpmf(dpmfsFileName, "concept1GivenC2C3C4")
##assert toolkit.testProb2(concept1DpmfC2C3C4), "Sum of probabilities should be always 1."
##
##concept1GivenC2C3C4 = gmtk.combineDtDcpt2(concept1DtC2C3C4, concept1DpmfC2C3C4)
concept1GivenC2C3C4 = jointProbC1C2C3C4.conditionalize([1, 2, 3])
##if verbose:
def main(self, conceptMap, symbolMap, symbolDt, dcpts, dpmfs, symbolName,
outDir):
conceptMap = LexMap().read(conceptMap)
symbolMap = LexMap().read(symbolMap)
_sink_ = int(symbolMap['_sink_'])
_SINK_ = int(conceptMap['_SINK_'])
conceptCard = len(conceptMap)
symbolCard = len(symbolMap)
jointDtC1C2C3C4 = gmtk.readDt(symbolDt)
jointDcptC1C2C3C4 = gmtk.readDcpt(dcpts, "jointProbC1C2C3C4")
jointProbC1C2C3C4 = gmtk.combineDtDcpt1(jointDtC1C2C3C4,
jointDcptC1C2C3C4)
symbolDpmfC1C2C3C4 = gmtk.readDpmf(dpmfs,
"%sGivenC1C2C3C4" % symbolName)
symbolGivenC1C2C3C4 = gmtk.combineDtDcpt2(jointDtC1C2C3C4,
symbolDpmfC1C2C3C4)
jointProbSymbolC1C2C3C4 = symbolGivenC1C2C3C4.multiple(
[1, 2, 3, 4], jointProbC1C2C3C4)
gmtk.saveDpmfsProbs(outDir, "%sGivenC1C2C3C4" % symbolName,
len(symbolGivenC1C2C3C4.vectSubList([1, 2, 3, 4])),
symbolCard, symbolGivenC1C2C3C4)
jointProbSymbolC1C2C3 = jointProbSymbolC1C2C3C4.marginalize(
[0, 1, 2, 3])
symbolGivenC1C2C3 = jointProbSymbolC1C2C3.conditionalize([1, 2, 3])
gmtk.saveDpmfsProbs(outDir, "%sGivenC1C2C3" % symbolName,
len(symbolGivenC1C2C3.vectSubList([1, 2, 3])),
symbolCard, symbolGivenC1C2C3)
jointProbSymbolC1C2 = jointProbSymbolC1C2C3.marginalize([0, 1, 2])
symbolGivenC1C2 = jointProbSymbolC1C2.conditionalize([1, 2])
gmtk.saveDpmfsProbs(outDir, "%sGivenC1C2" % symbolName,
len(symbolGivenC1C2.vectSubList([1, 2])),
symbolCard, symbolGivenC1C2)
jointProbSymbolC1 = jointProbSymbolC1C2.marginalize([0, 1])
symbolGivenC1 = jointProbSymbolC1.conditionalize([1])
# in case of conditioning by _SINK_ I have to enable to generate _sink_ word only
# otherwise I would see _SINK_ concept in the stack
# TODO: Turn it into validator
symbolGivenC1.setValue([_sink_, _SINK_], 1)
gmtk.saveDcptBigram(outDir, "%sGivenC1" % symbolName, symbolCard,
conceptCard, symbolGivenC1)
symbolUnigram = jointProbSymbolC1.marginalize([0])
# I need to enable to decode _unseen_ word only ! so set the probability of
# generating _empty_ to zero
symbolUnigram.setValue([int(symbolMap["_empty_"])], 0)
# normalize sum of probabilities to one
symbolUnigram = symbolUnigram.normJoint()
gmtk.saveDcptUnigram(outDir, "%sUnigram" % symbolName, symbolCard,
symbolUnigram)
gmtk.saveDcptUnseen(outDir, "%sZerogram" % symbolName, symbolCard,
symbolMap)
