import pandas as pd
import numpy as np
from numpy import *
import os
from surprise import Dataset
from surprise import NormalPredictor
from surprise import Reader
from surprise.model_selection import cross_validate
from surprise import KNNBaseline
from surprise.model_selection import cross_validate
from surprise.model_selection import GridSearchCV

class Recommandation(object):
    
    def __init__(self, filepath, k,user):
        self.filepath = filepath
        self.k = k
        self.id = user
    
    def getRecomMat(self):
        '''
        构造数据集
        '''
        
        file_path = os.path.expanduser(self.filepath)
        reader = Reader(line_format='user item rating timestamp', sep=',')
        user_data = Dataset.load_from_file(file_path, reader=reader)
        print("构建数据集...")
        trainset = user_data.build_full_trainset()
        '''
        训练模型
        '''
        # KNN-Baseline
        print("开始训练模型...")
        sim_options = {'name': 'pearson_baseline', 'user_based': False}
        #sim_options = {'name': 'cosine', 'user_based': False}
        algo1 = KNNBaseline(sim_options=sim_options)
        ans_test = algo1.fit(trainset)
        #cross_validate(algo1, user_data, measures=['RMSE', 'MAE'], cv=3, verbose=True)

        '''
        取出相似矩阵
        '''
        original_sim_matrix = ans_test.sim.copy()

        '''
        计算推荐矩阵
        '''
        # 转化为矩阵
        sim_matrix = mat(original_sim_matrix) # 这里的相似矩阵是对应trainset的index

        # 用户做题记录 1为写了 0为未写
        size_item = len(original_sim_matrix[0])
        user_inner_id = []
        item_inner_id = []
        rating_inner = []
        for (u, i, r) in trainset.all_ratings():
            #print(u, i, r)
            user_inner_id.append(u)
            item_inner_id.append(i)
            rating_inner.append(r)
        size_user = len(set(user_inner_id))
        user_record = np.zeros((size_item,size_user))
        n = len(user_inner_id)                    
        for i in range(0, n) :
            index_i = item_inner_id[i]
            index_j = user_inner_id[i]
            user_record[index_i][index_j] = 1   
        user_submit_record = mat(user_record)

        # 生成每个用户所做的所有题目与第a道的相似度之和
        user_with_all_sim = dot(sim_matrix, user_submit_record)

        # 生成每个题目和所有题目的所有相似度之和
        one_maxtrix = mat(ones((size_item,size_user)))
        item_with_all_sim = dot(sim_matrix,one_maxtrix)

        # 处理item_with_all_sim的0元素
        item_with_all_sim_1 = item_with_all_sim.getA().copy()
        for i in range(0,size_item):
            for j in range(0, size_user):
                if item_with_all_sim_1[i][j]==0:
                    item_with_all_sim_1[i][j] = -Inf
        item_with_all_sim_1

        # 计算加权平均， 生成推荐矩阵
        recommand_matrix = user_with_all_sim / item_with_all_sim_1

        '''
        对推荐矩阵进行处理
        '''
        # 将推荐矩阵中用户做过的题目设置为-Tnf
        recommand_arr = recommand_matrix.getA().copy()
        user_submit_arr = user_record.copy()
        size_raw = len(user_record)
        size_col = len(user_record[0])
        for i in range(0,size_raw):
            for j in range(0, size_col):
                if user_submit_arr[i][j]==1 :
                    recommand_arr[i][j] = -Inf

        # 排序
        test_arr = recommand_arr.copy()
        index_arr = np.argsort(-test_arr, axis = 0) #按列排序

        '''
        计算最后的top-k推荐矩阵
        '''
        full_recom_arr = (recommand_arr.copy()).T
        recon_index = index_arr.copy().T
        size_top = 10
        recom_topk = []
        tmp_topk = []

        for i in range(0, size_user): # 用户
            tmp_topk = [] # 保存用id + 推荐前20的题目id
            user = algo1.trainset.to_raw_uid(i) # 用户
            tmp_topk.append(user)
            for j in range(0, size_top): # 题目
                index_j = recon_index[i][j]  # 这里的index就是item内部的id
                problem_inner_id = algo1.trainset.to_raw_iid(index_j)
                tmp_topk.append(problem_inner_id)
            recom_topk.append(tmp_topk)

        
        # 将推荐矩阵写入文件
        print('将推荐矩阵写入文件...')
        df_recom_top = pd.DataFrame(recom_topk)
#        print(df_recom_top.icol[0])
        return df_recom_top[df_recom_top.iloc[:,0]==self.id]
#        print(df_recom_top[df_recom_top[:0]=='*Miracle*'])
#        df_recom_top.to_csv('output.txt', encoding='utf-8', index=False, header=False)


# 调用方式

test_path = 'new_suprise_data.txt'
test = Recommandation(test_path, 10, "wgy")
test.getRecomMat()