kNN.py

from math import sqrt

import numpy as np

def cosine(dataA, dataB):
    if type(dataA) is list and type(dataB) is list:
        if len(dataA) != len(dataB):
            print("Not same length!")
            return -1

        dataLength = len(dataA)
        AB = sum([dataA[i]*dataB[i] for i in range(dataLength)])
        normA = sqrt(sum([i**2 for i in dataA]))
        normB = sqrt(sum([i**2 for i in dataB]))

        denominator = normA * normB
        if denominator == 0:
            return 0
        return AB/denominator
    else:
        print("Input data invalid!")
        return -1

def pearson(dataA, dataB):
    if type(dataA) is list and type(dataB) is list:
        if len(dataA) != len(dataB):
            print("Not same length!")
            return -1

        dataLength = len(dataA)
        intersection = [i for i in range(dataLength) if dataA[i]!=0 and dataB[i]!=0]
        #intersection = [i for i in range(dataLength)]
        if len(intersection) == 0:
            return 0

        avgA = np.mean([i for i in dataA if i != 0])
        avgB = np.mean([i for i in dataB if i != 0])
        #avgA = np.mean([i for i in dataA])
        #avgB = np.mean([i for i in dataB])
        numerator = sum([(dataA[i] - avgA)*(dataB[i] - avgB) for i in intersection])
        deviationA = sqrt(sum([(dataA[i]-avgA)**2 for i in intersection]))
        deviationB = sqrt(sum([(dataB[i]-avgB)**2 for i in intersection]))
        if(deviationA * deviationB) == 0:
            return 0
        return numerator/(deviationA*deviationB)


def kNN(data, measure, k = None):
    #simulatedData = [[round(np.corrcoef(i,j)[0,1],3) for j in data] for i in data]
    simulatedData = [[round(measure(i,j),3) for j in data] for i in data]
    #'''
    print("Similarity Data")
    for i in simulatedData:
        print(i)
    #'''
    print(k)
    newData = []
    for i in range(len(data)):
        newData.append([])
        for j in range(len(data[i])):
            if(data[i][j] == 0):
                #print("i,j = ", i,j)
                userCol = [(simulatedData[i][index], index) for index in range(len(simulatedData[i]))]
                #print("userCol")
                userCol.sort(reverse = True)
                #print(userCol)
                #taken = userCol[1:k+1]
                taken = []
                countk = 0
                #print("k:",k)
                for l in range(0,len(userCol)):
                #    print(l)
                #    print("data: ",data[userCol[l][1]][j])
                    if userCol[l][1] == i:
                        continue
                    if (data[userCol[l][1]][j] != 0):
                        countk += 1
                        taken.append(userCol[l])
                    if (countk == k):
                        #print("noooowat")
                        break
                
                #print("taken")
                #print(taken)
                #print(data)
                
                #k = len(taken)
                predictingMean = np.mean([index for index in data[i] if index != 0])
                similarUserMean = [np.mean([l for l in data[index[1]] if l != 0]) for index in taken]
                
                #print("SimilarUserMean")
                #print(similarUserMean)
                #print([data[i][taken[k][1]] for k in range(len(taken))])
                #print()
                #print(simulatedData[i][taken[k][1]], data[taken[k][1]][j], similarUserMean[k])
                numerator = sum([simulatedData[i][taken[index][1]]*(data[taken[index][1]][j] - similarUserMean[index]) for index in range(len(taken))])
                #print(predictingMean)
                denominator = sum([simulatedData[i][taken[index][1]] for index in range(len(taken))])
                #print(denominator)
                #print([simulatedData[i][taken[k][1]] for k in range(len(taken))])
                #print()
                if (denominator == 0):
                    newData[i].append(0)
                    continue
                res = predictingMean + numerator/denominator
                newData[i].append(round(res,3))
            else:
                newData[i].append(data[i][j])
    return newData


'''
from scipy import spatial

table2 = [[round(1 - spatial.distance.cosine(i, j),3) for j in data] for i in data]
print()
for i in table2:
    print(i)

table3 = [[round(pearson(i,j),3) for j in data] for i in data]
print()
for i in table3:
    print(i)

table4 = [[np.corrcoef(i,j)[0,1] for j in data] for i in data]
print()
for i in table4:
    print(i)
'''

def readFile(fileString):
    return np.loadtxt(fileString, delimiter=" ", dtype = "int").tolist()

if __name__ == "__main__":
    fileString = r'F:\Tut\Sem6\Information System\testData\test3.txt'
    data = readFile(fileString)
    print("DATA:")
    for i in data:
        print(i)

    print("KNN Pearson:")
    blah = kNN(data, pearson, 2)
    for i in blah:
        print(i)
    
    print("\nKNN Cosine")
    bleh = kNN(data, cosine, 2)
    for i in bleh:
        print(i)