def __init__(self,type,rootDir):
"""
type: 1 -> full data
2 -> limited data
"""
#percentage of data to be used for model build
if type == 1:
self.GROUPTYPE = ["GENERAL"]
self.percentageList = [1000, 2500, 5000, 7500, 10000, 20000]
elif type == 2:
self.GROUPTYPE = ["GENERAL"]
self.percentageList = [1000]
else:
sys.exit("Wrong 'type' parameter in createData.__init__")
self.shevaDB = ShevaDB()
self.shevaTPF = ShevaTPF()
self.shevaUtils = ShevaUtils()
self.shevaVect = ShevaVect()
if rootDir != "":
self.shevaUtils.createDirOne(str(rootDir))
self.rootDir = str(rootDir)
else:
sys.exit("Wrong 'rootDir' parameter in createData.__init__")
def __init__(self, testSize, category, type=1):
"""
INPUT:
type = full data (1) or training data (2)
testSize = % of model size (nr of documents in model) to test with
"""
#percentage of data to be used for model build
if type == 1:
self.GROUPTYPE = ["CATID", "FATHERID", "GENERAL"]
self.limitList = [1000, 2500, 5000, 7500, 10000, 20000]
elif type == 2:
self.GROUPTYPE = ["GENERAL", "FATHERID", "GENERAL"]
self.limitList = [1000]
else:
sys.exit("Wrong 'type' parameter in createData.__init__")
print "SimilarityLevel created"
#Sheva Objects
self.shevaDB = ShevaDB()
self.shevaTPF = ShevaTPF()
self.shevaSimilarity = ShevaSimilarity()
self.shevaCSV = ShevaCSV()
self.shevaClassificationMetrics = ShevaClassificationMetrics()
self.shevaUtils = ShevaUtils()
#SimilarityLevel Variables
self.rootDir = "LimitModels/"
self.testSize = testSize
self.category = category
self.maxDepth = self.shevaDB.getCategorymaxDepth(self.category)
def __init__(self, testSize):
"""
INPUT:
type = full data (1) or training data (2)
testSize = % of model size (nr of documents in model) to test with
"""
"""
#percentage of data to be used for model build
if type == 1:
self.GROUPTYPE = ["CATID", "FATHERID", "GENERAL"]
#self.percentageList = [25, 50, 75, 100]
elif type == 2:
self.GROUPTYPE = ["GENERAL"]
#self.percentageList = [25, 50, 75, 100]
else:
sys.exit("Wrong 'type' parameter in createData.__init__")
"""
print "SimilarityLevel created"
#Sheva Objects
self.shevaDB = ShevaDB()
self.shevaTPF = ShevaTPF()
self.shevaSimilarity = ShevaSimilarity()
self.shevaCSV = ShevaCSV()
self.shevaClassificationMetrics = ShevaClassificationMetrics()
self.shevaUtils = ShevaUtils()
#SimilarityLevel Variables
self.rootDir = "LevelModels/"
self.testSize = testSize
class createDataSingleLimit:
def __init__(self,type,rootDir):
"""
type: 1 -> full data
2 -> limited data
"""
#percentage of data to be used for model build
if type == 1:
self.GROUPTYPE = ["GENERAL"]
self.percentageList = [1000, 2500, 5000, 7500, 10000, 20000]
elif type == 2:
self.GROUPTYPE = ["GENERAL"]
self.percentageList = [1000]
else:
sys.exit("Wrong 'type' parameter in createData.__init__")
self.shevaDB = ShevaDB()
self.shevaTPF = ShevaTPF()
self.shevaUtils = ShevaUtils()
self.shevaVect = ShevaVect()
if rootDir != "":
self.shevaUtils.createDirOne(str(rootDir))
self.rootDir = str(rootDir)
else:
sys.exit("Wrong 'rootDir' parameter in createData.__init__")
def createData(self, category):
"""
1. get root categories to be used and iterate through main categories
3. get max depth for individual category
4. from 1 to max till 1 to 1
get all catid for iterated category
get all pages for selected categories
call createCorpusAndVectorModel from selected documents
"""
for group in self.GROUPTYPE:
#gruping dependent queries
if group != "FATHERID":
sqlCategory = "select Description, catid from dmoz_combined where mainCategory = '%s' limit 20000" %(category)
else:
sqlCategory = "select Description, fatherid from dmoz_combined where mainCategory = '%s' limit 20000" %(category)
sqlQueryResults = self.shevaDB.dbQuery(sqlCategory)
if sqlQueryResults == 0:
sys.exit("SQL code error in level: \t", category,"\t",indeks,"\t",sqlCategoryLevel)
for percentageItem in self.percentageList:
sqlQueryResultsLimit = [x for x in sqlQueryResults[:percentageItem]]
#data for % model, range data
dataCategoryLevelAll = []
dataCategoryLabelAll = []
originalCatIDAll = []
dataCategorySingleAll = []
path = "%s/%s/%s/" %(self.rootDir,group,percentageItem)
self.shevaUtils.createDir(self.rootDir,group,percentageItem)
#for indeks in ranger:
#level list variables
dataCategoryLevel = []
dataCategoryLabel = []
originalCatID = []
originalFatherID = []
finalContent = []
#get unique values
if group == "GENERAL":
finalContent = [[item for item in row[0].split()] for row in sqlQueryResultsLimit]
originalCatID = [row[1] for row in sqlQueryResultsLimit]
dataCategoryLevel.extend(self.shevaTPF.returnClean(finalContent,1))
else:
unique = []
for row in sqlQueryResultsLimit:
if row[1] not in unique:
unique.append(row[1])
#prepare rows with uniq for document in model
for uniq in unique:
tempUnique = []
tempUnique = [row[0].split() for row in sqlQueryResultsLimit if row[1] == uniq]
mergedContent = [i for i in itertools.chain.from_iterable(tempUnique)]
finalContent.append(mergedContent)
originalCatID.append(uniq)
dataCategoryLevel.extend(self.shevaTPF.returnClean(finalContent,1))
self.shevaUtils.createDir(self.rootDir, group, percentageItem)
#create file names
#fileNameAll = "%s_%s_1_%s" %(str(percentageItem),category,str(indeks))
fileNameLevel = "%s_%s" %(str(percentageItem),category)
fileNameSingleAll = "%s_%s" %(str(percentageItem),category)
########## ORIGINAL DESCRIPTION AND VECTORIZATION #################
#create corpus models
#self.shevaVect.createCorpusAndVectorModel(dataCategoryLevel,fileNameLevel,path)
dataCategoryLevelAll.extend(dataCategoryLevel)
#self.shevaVect.createCorpusAndVectorModel(dataCategoryLevelAll, fileNameAll,path)
#single model for all documents
dataCategorySingleAll.append([x for sublist in dataCategoryLevelAll for x in sublist])
#print dataCategorySingleAll
#print len(Counter(dataCategorySingleAll[0]))
print len(dataCategorySingleAll), len(dataCategorySingleAll[0])
self.shevaVect.createCorpusAndVectorModel(dataCategorySingleAll, fileNameSingleAll,path)
########## ORIGINAL CATEGORIES ID #################
#self.shevaUtils.getCategoryListLevel(originalCatID,fileNameLevel,path)
#originalCatIDAll.extend(originalCatID)
#self.shevaUtils.getCategoryListLevel(originalCatIDAll,fileNameAll,path)
#print out number of documents for (cat,level,model)
print "Done with:\t",group,"\t",category,"\t","\t",percentageItem
class SimilarityLimit:
##@profile w
def __init__(self, testSize, category, type=1):
"""
INPUT:
type = full data (1) or training data (2)
testSize = % of model size (nr of documents in model) to test with
"""
#percentage of data to be used for model build
if type == 1:
self.GROUPTYPE = ["CATID", "FATHERID", "GENERAL"]
self.limitList = [1000, 2500, 5000, 7500, 10000, 20000]
elif type == 2:
self.GROUPTYPE = ["GENERAL", "FATHERID", "GENERAL"]
self.limitList = [1000]
else:
sys.exit("Wrong 'type' parameter in createData.__init__")
print "SimilarityLevel created"
#Sheva Objects
self.shevaDB = ShevaDB()
self.shevaTPF = ShevaTPF()
self.shevaSimilarity = ShevaSimilarity()
self.shevaCSV = ShevaCSV()
self.shevaClassificationMetrics = ShevaClassificationMetrics()
self.shevaUtils = ShevaUtils()
#SimilarityLevel Variables
self.rootDir = "LimitModels/"
self.testSize = testSize
self.category = category
self.maxDepth = self.shevaDB.getCategorymaxDepth(self.category)
def __del__(self):
print 'SimilarityLevel destroyed'
#@profile
def calculateLimitSimilarity(self):
for limit in self.limitList:
for group in self.GROUPTYPE:
print "####################################################################"
#print category, group, percentage, debth
sim = []
vec_bow = []
allCategoryDataOID = []
categoryDataOID = []
categoryData = []
print "created variables"
#path & csv file
path = "%s%s/%s/" %(self.rootDir,group,limit)
fileName = "%s_%s" %(limit,self.category)
IODfilePath = "%soriginalID/%s.csv" %(path,fileName)
print "Setup paths"
#get data from original ID csv; unique ID
allCategoryDataOID = self.shevaCSV.getModelCSV(IODfilePath)
#categoryDataOID = self.shevaCSV.getIDfromModel(IODfilePath)
print "Got all modelRow->originalID mappings"
#get sim index, model, dict
indexDir = "%sindexFiles/" %(path)
self.shevaUtils.createDirOne(indexDir)
index, tfidfModel, dictionary, corpusSize = self.shevaSimilarity.getSimilarityIndex(path, fileName, group)
#return sample from original data
categoryDataOID, categoryData = self.shevaDB.getSample(limit,self.testSize,self.category,self.maxDepth, group)
#calculate similarites
cleanText = self.shevaTPF.returnClean(categoryData, 1)
cleanTextBoW = [dictionary.doc2bow(cleanText[i]) for i in range(0, len(cleanText))]
print "Done with bow representation"
vec_bow = self.shevaSimilarity.convert2VSM(cleanTextBoW, tfidfModel)
print len(vec_bow)
simCalculation = self.shevaSimilarity.calculateSimilarity(index, vec_bow, 0.1)
#calcualte IR measures
cPrecision, cRecall, cF1 = self.shevaClassificationMetrics.computeClassificationMetrics(categoryDataOID, allCategoryDataOID, simCalculation)
print "All data measures :\t\t\t\tPrecision:\t", cPrecision, "\t\tRecall\t", cRecall, "\t\tF1:\t", cF1
sqlClassic = "INSERT INTO analysis_results_limit (category, groupingType, limitValue, levelDepth, testSize, measureType, P, R, F1) VALUES ('%s', '%s', '%s', '%i', '%i', '%s','%f','%f','%f')" % (self.category,group,limit,self.maxDepth,self.testSize, "computeClassificationMetrics",cPrecision, cRecall, cF1)
self.shevaDB.dbQuery(sqlClassic)
cPrecisionR, cRecallR, cF1R = self.shevaClassificationMetrics.computeClassificationMetricsRelative(categoryDataOID, allCategoryDataOID, simCalculation)
print "Relative (with or) data measures :\t\tPrecision:\t", cPrecisionR, "\t\tRecall\t", cRecallR, "\t\tF1:\t", cF1R
sqlRelative = "INSERT INTO analysis_results_limit (category, groupingType, limitValue, levelDepth, testSize,measureType, P, R, F1) VALUES ('%s', '%s', '%s', '%i', '%i', '%s','%f','%f','%f')" % (self.category,group,limit,self.maxDepth,self.testSize, "computeClassificationMetricsRelative",cPrecisionR, cRecallR, cF1R)
self.shevaDB.dbQuery(sqlRelative)
cPrecisionE, cRecallE, cF1E = self.shevaClassificationMetrics.computeClassificationMetricsExclusive(categoryDataOID, allCategoryDataOID, simCalculation)
print "Exclusive (with and) data measures :\t\tPrecision:\t", cPrecisionE, "\t\tRecall\t", cRecallE, "\t\tF1:\t", cF1E
sqlExclusive = "INSERT INTO analysis_results_limit (category, groupingType, limitValue, levelDepth,testSize, measureType, P, R, F1) VALUES ('%s', '%s', '%s', '%i', '%i', '%s','%f','%f','%f')" % (self.category,group,limit,self.maxDepth,self.testSize, "computeClassificationMetricsExclusive",cPrecisionE, cRecallE, cF1E)
self.shevaDB.dbQuery(sqlExclusive)
#trying to figure out the memory thing. needs speed-up in performance otherwise...
dbData = []
simCalculation = []
cleanText = []
cleanTextBoW = []
vec_bow = []
del index
del tfidfModel
del dictionary
del corpusSize
del simCalculation
del vec_bow
del allCategoryDataOID
del categoryDataOID
del categoryData
del dbData
gc.collect()
class createDataSingleLevel:
def __init__(self,type,rootDir):
"""
type: 1 -> full data
2 -> limited data
"""
#percentage of data to be used for model build
if type == 1:
self.GROUPTYPE = ["CATID","FATHERID","GENERAL"]
self.percentageList = [25, 50, 75, 100]
elif type == 2:
self.GROUPTYPE = ["GENERAL"]
self.percentageList = [5]
else:
sys.exit("Wrong 'type' parameter in createData.__init__")
self.shevaDB = ShevaDB()
self.shevaTPF = ShevaTPF()
self.shevaUtils = ShevaUtils()
self.shevaVect = ShevaVect()
self.shevaCSV = ShevaCSV()
if rootDir != "":
self.shevaUtils.createDirOne(str(rootDir))
self.rootDir = str(rootDir)
else:
sys.exit("Wrong 'rootDir' parameter in createData.__init__")
def createData(self, category):
"""
1. get root categories to be used and iterate through main categories
3. get max depth for individual category
4. from 1 to max till 1 to 1
get all catid for iterated category
get all pages for selected categories
call createCorpusAndVectorModel fro selected documents
"""
ranger = self.shevaDB.getCategoryDepth(category)
for group in self.GROUPTYPE:
sqlQueryResults = []
#gruping dependent queries
if group == "FATHERID":
sqlCategory = "select Description, fatherid, categoryDepth from dmoz_combined where mainCategory = '%s'" %(category)
else:
sqlCategory = "select Description, catid, categoryDepth from dmoz_combined where mainCategory = '%s'" %(category)
sqlQueryResults = self.shevaDB.dbQuery(sqlCategory)
if sqlQueryResults == 0:
sys.exit("SQL code error in level: \t", category,"\t",indeks,"\t",sqlCategoryLevel)
for percentageItem in self.percentageList:
#data for % model, range data
dataCategoryLevelAll = []
dataCategoryLabelAll = []
originalCatIDAll = []
dataCategorySingleAll = [[]]
path = "%s/%s/%s/" %(self.rootDir,group,percentageItem)
self.shevaUtils.createDir(self.rootDir,group,percentageItem)
#var = ""
for indeks in ranger:
#var += "Level:\t%s\n" %(indeks)
"""
#gruping dependent queries
if group == "FATHERID":
sqlCategoryLevel = "select Description, fatherid from dmoz_combined where mainCategory = '%s' and categoryDepth = '%s'" %(category,indeks)
else:
sqlCategoryLevel = "select Description, catid from dmoz_combined where mainCategory = '%s' and categoryDepth = '%s'" %(category,indeks)
sqlQueryResultsLevel = self.shevaDB.dbQuery(sqlCategoryLevel)
"""
sqlQueryResultsLevel = [x for x in sqlQueryResults if x[2] == indeks]
#level list variables
finalContent = []
dataCategoryLevel = []
dataCategoryLabel = []
originalCatID = []
#originalFatherID = []
#get unique values
if group == "GENERAL":
#finalContent = []
percentageLevel = self.shevaUtils.setLimit(percentageItem,sqlQueryResultsLevel)
finalContent = [[item for item in row[0].split()] for row in sqlQueryResultsLevel[:percentageLevel]]
#var += "Original words:\t%s\n" %(finalContent)
originalCatID = [row[1] for row in sqlQueryResultsLevel[:percentageLevel]]
#var += "Original IDs:\t%s\n" %(originalCatID)
dataCategoryLevel.extend(self.shevaTPF.returnClean(finalContent,1))
#var += "Processed words:\t%s\n" %(dataCategoryLevel)
else:
unique = []
for row in sqlQueryResultsLevel:
if row[1] not in unique:
unique.append(row[1])
for uniq in unique:
#var += "ID:\t%s\n" %(uniq)
tempUnique = []
tempUnique = [row[0] for row in sqlQueryResultsLevel if row[1] == uniq]
percentageLevel = self.shevaUtils.setLimit(percentageItem,tempUnique)
tempUnique = [item.split() for item in tempUnique[:percentageLevel]]
mergedContent = [i for i in itertools.chain.from_iterable(tempUnique)]
#var += "Original words:\t%s\n" %(mergedContent)
finalContent.append(mergedContent)
originalCatID.append(uniq)
dataCategoryLevel.extend(self.shevaTPF.returnClean(finalContent,1))
self.shevaUtils.createDir(self.rootDir, group, percentageItem)
########## FILE NAMES #################
fileNameAll = "%s_%s_1_%s" %(str(percentageItem),category,str(indeks))
fileNameLevel = "%s_%s_%s" %(str(percentageItem),category,str(indeks))
fileNameSingleAll = "%s_%s_%s" %(str(percentageItem),category,str(indeks))
########## PRINT OUT ORIGINAL AND PROCESSED DATA #################
"""
print originalCatID
print finalContent
print dataCategoryLevel
"""
########## ORIGINAL DESCRIPTION AND VECTORIZATION #################
#self.shevaVect.createCorpusAndVectorModel(dataCategoryLevel,fileNameLevel,path)
dataCategoryLevelAll.extend(dataCategoryLevel)
#self.shevaVect.createCorpusAndVectorModel(dataCategoryLevelAll, fileNameAll,path)
#single model for all documents
dataCategorySingleAll[0].extend([x for sublist in dataCategoryLevelAll for x in sublist])
self.shevaVect.createCorpusAndVectorModel(dataCategorySingleAll, fileNameSingleAll, path)
########## ORIGINAL CATEGORIES ID #################
#self.shevaUtils.getCategoryListLevel(originalCatID,fileNameLevel,path)
originalCatIDAll.extend(originalCatID)
#self.shevaCSV.getCategoryListLevel(originalCatIDAll,fileNameAll,path)
#print out number of documents for (cat,level,model)
print "Done with:\t",group,"\t",category,"\t",indeks,"\t",percentageItem
####################### GC #################
del dataCategoryLevel
del originalCatID
gc.collect()
del dataCategoryLevelAll
del dataCategoryLabelAll
del originalCatIDAll
del dataCategorySingleAll
del sqlQueryResults
gc.collect()