recommendations

import pandas as pd
import numpy as np

file_obj = open('coursera_sessions_train.txt', 'r')
data_train = list(file_obj)


file_obj2 = open('coursera_sessions_test.txt', 'r')
data_test = list(file_obj2)

views_freq = dict()
buys_freq = dict()

for line in data_train:
    views, buys = line.strip().split(';')
    views = views.split(',')
    buys = buys.split(',')
    for id_viewed in views:
        if id_viewed in views_freq:
            views_freq[str(id_viewed)] += 1
        else:
            views_freq[str(id_viewed)] = 1

    for id_bought in buys:
        if id_bought in buys_freq:
            buys_freq[str(id_bought)] += 1
        else:
            buys_freq[str(id_bought)] = 1

# 1. Сортировка по просмотрам, на тренировочных данных


prec_1_list = []
prec_5_list = []
rec_1_list = []
rec_5_list = []

for line in data_train:
    views, buys = line.strip().split(';')
    views = views.split(',')
    buys = buys.split(',')
    if buys[0] != '':

        views = np.unique(np.array(views))

        freqs=[]
        views = list(views)
        for i in range(len(views)):
            freqs.append([i, views[i], views_freq[str(views[i])]])


        freqs.sort(cmp=lambda (x1, y1, z1), (x2, y2, z2): cmp(-z1, -z2) or cmp(x1, x2))

        #для AP@1, AR@1
        recom_1 = freqs[0][1]
        if recom_1 in buys:
            prec_1 = 1
            rec_1 = 1. / len(buys)
        else:
            prec_1 = 0
            rec_1 = 0
        prec_1_list.append(prec_1)
        rec_1_list.append(rec_1)

        #для AP@5, AR@5
        recom_5 = []
        if len(views) < 5:
            recom_5 = views
            znam = 0
            for j in range(len(views)):
                if recom_5[j] in buys:
                    znam += 1
            #prec_5 = float(znam) / len(views)
            prec_5 = float(znam) / 5
            rec_5 = float(znam) / len(buys)

        else:
            znam = 0
            for j in range(5):
                recom_5.append(freqs[j][1])
                if recom_5[j] in buys:
                    znam += 1
            prec_5 = float(znam) / 5
            rec_5 = float(znam) / len(buys)

        prec_5_list.append(prec_5)
        rec_5_list.append(rec_5)

ans1 = []
ans1.append(np.array(rec_1_list).mean())
ans1.append(np.array(prec_1_list).mean())
ans1.append(np.array(rec_5_list).mean())
ans1.append(np.array(prec_5_list).mean())

print ans1


#2. Сортировка по просмотрам, на тестовых данных

prec_1_list = []
prec_5_list = []
rec_1_list = []
rec_5_list = []

for line in data_test:
    views, buys = line.strip().split(';')
    views = views.split(',')
    buys = buys.split(',')
    if buys[0] != '':
        views = np.unique(np.array(views))

        freqs=[]
        views = list(views)
        for i in range(len(views)):
            if views[i] in views_freq:
                freqs.append([i, views[i], views_freq[str(views[i])]])
            else:
                freqs.append([i, views[i], 0])

        if freqs != []:
            freqs.sort(cmp=lambda (x1, y1, z1), (x2, y2, z2): cmp(-z1, -z2) or cmp(x1, x2))

        #для AP@1, AR@1
            recom_1 = freqs[0][1]
            if recom_1 in buys:
                prec_1 = 1
                rec_1 = 1. / len(buys)
            else:
                prec_1 = 0
                rec_1 = 0
            prec_1_list.append(prec_1)
            rec_1_list.append(rec_1)


            #для AP@5, AR@5
            #recom_5 = []
            znam = 0
            for j in range(5):
                if j < len(freqs):
                    #recom_5.append(freqs[j][1])
                    if freqs[j][1] in buys:
                        znam += 1
            prec_5 = float(znam) / 5
            rec_5 = float(znam) / len(buys)

            prec_5_list.append(prec_5)
            rec_5_list.append(rec_5)


ans2 = []
ans2.append(np.array(rec_1_list).mean())
ans2.append(np.array(prec_1_list).mean())
ans2.append(np.array(rec_5_list).mean())
ans2.append(np.array(prec_5_list).mean())

print ans2


# 3. Сортировка по покупкам, на тренировочных данных

prec_1_list = []
prec_5_list = []
rec_1_list = []
rec_5_list = []

for line in data_train:
    views, buys = line.strip().split(';')
    views = views.split(',')
    buys = buys.split(',')
    if buys[0] != '':
        views = np.unique(np.array(views))

        freqs=[]
        views = list(views)
        for i in range(len(views)):
            if views[i] in buys_freq:
                freqs.append([i, views[i], buys_freq[str(views[i])]])
            #else:
                #freqs.append([i, views[i], 0])

        #print freqs
        if freqs != []:
            freqs.sort(cmp=lambda (x1, y1, z1), (x2, y2, z2): cmp(-z1, -z2) or cmp(x1, x2))

        #для AP@1, AR@1
            recom_1 = freqs[0][1]
            if recom_1 in buys:
                prec_1 = 1
                rec_1 = 1. / len(buys)
            else:
                prec_1 = 0
                rec_1 = 0
            prec_1_list.append(prec_1)
            rec_1_list.append(rec_1)


            #для AP@5, AR@5
            recom_5 = []
            znam = 0
            for j in range(5):
                if j < len(freqs):
                    recom_5.append(freqs[j][1])
                    if freqs[j][1] in buys:
                        znam += 1
            prec_5 = float(znam) / 5
            rec_5 = float(znam) / len(buys)

            prec_5_list.append(prec_5)
            rec_5_list.append(rec_5)


ans3 = []
ans3.append(np.array(rec_1_list).mean())
ans3.append(np.array(prec_1_list).mean())
ans3.append(np.array(rec_5_list).mean())
ans3.append(np.array(prec_5_list).mean())

print ans3


# 4. Сортировка по покупкам, на тестовых данных


prec_1_list = []
prec_5_list = []
rec_1_list = []
rec_5_list = []

count = 0
for line in data_test:

    views, buys = line.strip().split(';')
    views = views.split(',')
    buys = buys.split(',')
    if buys[0] != '':
        count += 1
        views = np.unique(np.array(views))
        buys = np.unique(np.array(buys))

        freqs=[]
        views = list(views)
        for i in range(len(views)):
            if views[i] in buys_freq:
                freqs.append([i, views[i], buys_freq[str(views[i])]])
            else:
                if views[i] not in views_freq:
                    freqs.append([i, views[i], 0])

        if freqs != []:
            freqs.sort(cmp=lambda (x1, y1, z1), (x2, y2, z2): cmp(-z1, -z2) or cmp(x1, x2))
            #freqs.sort(key = lambda (x1,y1,z1): -z1)

        #для AP@1, AR@1
            recom_1 = freqs[0][1]
            if recom_1 in buys:
                prec_1 = 1.
                rec_1 = 1. / len(buys)
            else:
                prec_1 = 0.
                rec_1 = 0.
            prec_1_list.append(prec_1)
            rec_1_list.append(rec_1)


        #для AP@5, AR@5
            recom_5 = []
            znam = 0.
            for j in range(5):
                if j < len(freqs):
                    recom_5.append(freqs[j][1])
                    if freqs[j][1] in buys:
                        znam += 1
            prec_5 = float(znam) / 5
            rec_5 = float(znam) / len(buys)

            prec_5_list.append(prec_5)
            rec_5_list.append(rec_5)


ans4 = []
ans4.append(np.array(rec_1_list).mean())
ans4.append(np.array(prec_1_list).mean())
ans4.append(np.array(rec_5_list).mean())
ans4.append(np.array(prec_5_list).mean())

print ans4