def wKNN(filename,k):
NUM_OF_DOCS = DocClass.objects.count()
FREQ_THRESHOLD = NUM_OF_DOCS*FREQ_THRESHOLD_PERCENT
#get all rows with frequency more than threshold
docInstances = DocFreqTable.objects.filter(docFreq__lte = FREQ_THRESHOLD)
wordList = []
for inst in docInstances:
wordList.append(getASCII(inst.word))
#Now start reading the file and then parse it match if the keywords are present in the file
#print(wordList)
filePointer = open(filename,encoding="latin-1")
#Push the keywords in the file to a dictionary datastructure
fileKeyWords = {}
wordsInFile = filePointer.read().split()
cnt = 0
stopwords=loadStopWords(STOP_WORD_LIST_FILENAME)
for word in wordsInFile:
if getASCII(word) in stopwords.keys():
continue
fileKeyWords[getASCII(word)] = True
cnt += 1
if cnt == MAX_NUM_OF_WORDS_READ:
break
#Get the tupple for the current file
##print(fileKeyWords)
inputFileTupple = getTuple(wordList,fileKeyWords)
##print(inputFileTupple.list)
docListInstance = DocClass.objects.all()
docList = []
# to get back the class name for any given document (assuming the number of documnets can be stored in the main mememory)
docToClassName = {}
for inst in docListInstance:
docList.append(inst.docName)
docToClassName[inst.docName] = inst.className
numRows = len(docList)
numCols = len(wordList)
knnMat = [[0]*numCols for i in range(numRows)]
rowNumber = 0
for doc in docList:
docWordInstances = WordTable.objects.filter(docName=doc)
docKeyWords = {}
for wordInstance in docWordInstances:
docKeyWords[wordInstance.word] = True
oneRow = getTuple(wordList,docKeyWords).list
x = 0
while x<numCols:
knnMat[rowNumber][x] = oneRow[x]
x += 1
rowNumber += 1
#get all the distance from the doc
distanceClassList = []
i = 0
while i < numRows:
newTuple = Tuple()
newTuple.list = knnMat[i]
newTuple.count = numCols
dis = inputFileTupple.distance(newTuple)
distanceClassList.append((dis,docToClassName[docList[i]]))
i += 1
distanceClassList.sort()
#print(distanceClassList)
classCount = {}
cnt = 0
for ins in distanceClassList:
wt = 1/(ins[0]+0.01) #for removing the zero error
#print(str(ins[1])+' -> '+str(wt))
if ins[1] in classCount.keys():
classCount[ins[1]] += wt
else:
classCount[ins[1]] = wt
if cnt == k:
break;
cnt += 1
#print(str(classCount))
classCountList = []
for className in classCount.keys():
classCountList.append((-1*classCount[className],className))
classCountList.sort()
answer_list = []
len_classCountList = len(classCountList)
sum_rows = 0
i = 0
while i < len_classCountList:
sum_rows += classCountList[i][0]
i += 1
i = 0
while i < len_classCountList:
answer_list.append(((classCountList[i][0]/(sum_rows))*100,classCountList[i][1]))
i += 1
return answer_list
def NaiveBayes(file):
#Logging to a file
orig_stdout = sys.stdout
#f = open('out.txt', 'w')
#sys.stdout = f
#Extract tokens from doc
#1. Select the first 300 words which
#2. Contains only letters (helps in removing meaningless tokens)
#3. Must not be a stop word
fp = open(file,encoding="latin-1")
docCount = DocClass.objects.count()
#print("Number of documents: ",docCount)
W = fp.read().split()
cnt=0
processedW={}
stopwords=loadStopWords(STOP_WORD_LIST_FILENAME)
for w_ in W:
w=w_.lower()
w = get_my_string(w)
if not w.isalpha():
continue
if w in processedW.keys():
continue
if w in stopwords.keys():
continue
processedW[w]=1
cnt += 1
if cnt == MAX_NUM_OF_WORDS_READ:
break
#List of all (className: "className") key value pairs
C = DocClass.objects.all().values('className').distinct()
#number of classes
countC=len(C)
#print("Number of classes: ",countC)
#(probability,class) pairs
scores=[]
#number of terms
B=DocFreqTable.objects.all().count()
#print("Number of terms: ",B)
#Outer Loop: for every class
for c in C:
cName=c['className']
#print("Calculating probabilities for class=",cName)
#get all docs corresponding to current class
docsC = DocClass.objects.filter(className = cName).values("docName")
priorC = len(docsC)/docCount
prob = log10(priorC) #prob is the probability of the test document of belonging to clas
#number of docs of current class in which a term of vocabulary cumulative for all terms
denC = WordTable.objects.filter(docName__in=docsC).count()
#Inner Loop: for every term
for t in processedW.keys():
#number of docs of current class in which the current term appears
numC = WordTable.objects.filter(docName__in=docsC,word=t).select_related().count()
#print("Using term=",t)
#how much evidence does term provide that clas is the correct class
prob= prob + log10( (1+numC)/(denC+B) )
scores.append( (-prob,cName) )
scores.sort()
answer_list = []
#make the probabilities non negative again
for i in scores:
answer_list.append( ( -i[0], i[1] ) );
sys.stdout = orig_stdout
#f.close()
return answer_list
