fix personal rank笔记

recall/personalRank/algorithm_demo/personal_rank_base_python.py

@@ -0,0 +1,111 @@
+# -*- encoding: utf-8 -*-
+"""
+@author: Aramil
+@date: 2022/10/4 22:41
+@brief: personal-rank基于python实现
+"""
+import operator
+import json
+
+import pandas as pd
+
+
+class PersonalRankBasePython(object):
+    """
+        基于python实现personal_rank原始公式
+    """
+    def __init__(self):
+        self.train_file = "../../../data/ml-100k/ua.base"
+        self.test_file = "../../../data/ml-100k/ua.test"
+        self.personal_rank_save_file = "../../../model/personal_rank/pr_user_item_recommend_dict_origin_by_python.json"
+        self.alpha = 0.8
+        self.iter_num = 100
+        self.topN = 100
+        pass
+
+    def read_data(self):
+        """
+            读取personal_rank数据
+        """
+        train_data = pd.read_csv(self.train_file, sep='\t', engine='python',
+                                 names=['user_id', 'movie_id', 'rate', 'event_timestamp'])
+        test_data = pd.read_csv(self.test_file, sep='\t', engine='python',
+                                names=['user_id', 'movie_id', 'rate', 'event_timestamp'])
+        print('=' * 20, '训练数据集', '=' * 20)
+        print(train_data.head(5))
+        print('=' * 20, '测试数据集', '=' * 20)
+        print(test_data.head(5))
+        return train_data, test_data
+
+    def get_personal_rank_graph(self, train_data):
+        """
+        :param train_data: personal_rank训练数据
+        :return: 构造personal_rank"用户-物品"二分图
+        """
+        personal_rank_graph = dict()
+        train_user_set = set()
+        for index, row in train_data.iterrows():
+            user_id = int(row['user_id'])
+            item_id = int(row['movie_id'])
+            personal_rank_graph.setdefault(user_id, dict())
+            personal_rank_graph[user_id][item_id] = 1
+            personal_rank_graph.setdefault(item_id, dict())
+            personal_rank_graph[item_id][user_id] = 1
+            train_user_set.add(user_id)
+        return personal_rank_graph, train_user_set
+
+    def personal_rank(self, personal_rank_graph, train_user_set):
+        user_recommend_result = dict()
+        cnt = 0
+        print(len(list(train_user_set)))
+        for user_id in train_user_set:
+            rank = {point: 0 for point in personal_rank_graph}
+            rank[user_id] = 1
+            recommend_result = dict()
+            for iter_index in range(self.iter_num):
+                tmp_rank = {point: 0 for point in personal_rank_graph}
+                for out_point, out_dict in personal_rank_graph.items():
+                    for inner_point, value in personal_rank_graph[out_point].items():
+                        tmp_rank[inner_point] += round(self.alpha * rank[out_point]/len(out_dict), 4)
+                        if inner_point == user_id:
+                            tmp_rank[inner_point] += round(1 - self.alpha, 4)
+                if tmp_rank == rank:
+                    break
+                rank = tmp_rank
+            right_num = 0
+            for zuhe in sorted(rank.items(), key=operator.itemgetter(1), reverse=True):
+                point, pr_score = zuhe[0], zuhe[1]
+                if point in personal_rank_graph[user_id]:
+                    continue
+                recommend_result.setdefault(point, 0)
+                recommend_result[point] = pr_score
+                right_num += 1
+                if right_num > self.topN:
+                    break
+            user_recommend_result.setdefault(user_id, dict())
+            user_recommend_result[user_id] = recommend_result
+            cnt += 1
+            if cnt % 10 == 0:
+                print(cnt)
+        return user_recommend_result
+
+    def save_user_recommend_result_dict(self, user_recommend_result):
+        json.dump(user_recommend_result, open(self.personal_rank_save_file, 'w'))
+        print("pr original by python 保存成功！")
+        pass
+
+    def run(self):
+        train_data, test_data = self.read_data()
+        personal_rank_graph, train_user_set = self.get_personal_rank_graph(train_data)
+        user_recommend_result = self.personal_rank(personal_rank_graph, train_user_set)
+        self.save_user_recommend_result_dict(user_recommend_result)
+
+
+def main():
+    prbp = PersonalRankBasePython()
+    prbp.run()
+
+
+if __name__ == '__main__':
+    exit(main())
+

recall/personalRank/algorithm_demo/personal_rank_mat_base_python.py

@@ -0,0 +1,93 @@
+# -*- encoding: utf-8 -*-
+"""
+@author: Hanjingwei
+@date: 2022/10/13 10:26
+@brief: personal-rank矩阵式基于python实验
+"""
+import numpy as np
+from scipy.sparse import coo_matrix
+
+
+class PRMatBasePython(object):
+    def __init__(self):
+        self.train_file = "../../../data/ml-100k/ua.base"
+        self.test_file = "../../../data/ml-100k/ua.test"
+        self.personal_rank_save_file = "../../../model/personal_rank/pr_user_item_recommend_dict_mat_by_python.json"
+        self.alpha = 0.8
+        self.iter_num = 100
+        self.topN = 100
+
+    def read_data(self):
+        train_data = pd.read_csv(self.train_file, sep='\t', engine='python',
+                                 names=['user_id', 'movie_id', 'rate', 'event_timestamp'])
+        test_data = pd.read_csv(self.test_file, sep='\t', engine='python',
+                                names=['user_id', 'movie_id', 'rate', 'event_timestamp'])
+        print('=' * 20, '训练数据集', '=' * 20)
+        print(train_data.head(5))
+        print('=' * 20, '测试数据集', '=' * 20)
+        print(test_data.head(5))
+        return train_data, test_data
+
+    def get_personal_rank_graph(self, train_data):
+        """
+        :param train_data: personal_rank训练数据
+        :return: 构造personal_rank"用户-物品"二分图
+        """
+        personal_rank_graph = dict()
+        train_user_set = set()
+        for index, row in train_data.iterrows():
+            user_id = int(row['user_id'])
+            item_id = int(row['movie_id'])
+            personal_rank_graph.setdefault(user_id, dict())
+            personal_rank_graph[user_id][item_id] = 1
+            personal_rank_graph.setdefault(item_id, dict())
+            personal_rank_graph[item_id][user_id] = 1
+            train_user_set.add(user_id)
+        return personal_rank_graph, train_user_set
+
+    def graph_to_mat(self, personal_rank_graph):
+        vertex = list(personal_rank_graph.keys())
+        address_dict = {}
+        total_len = len(vertex)
+        for index in range(len(vertex)):
+            address_dict[vertex[index]] = index
+        row = []
+        col = []
+        data = []
+        for element_i in personal_rank_graph:
+            weight = round(1 / len(personal_rank_graph[element_i]), 3)
+            row_index = address_dict[element_i]
+            for element_j in personal_rank_graph[element_i]:
+                col_index = address_dict[element_j]
+                row.append(row_index)
+                col.append(col_index)
+                data.append(weight)
+        row = np.array(row)
+        col = np.array(col)
+        data = np.array(data)
+        m = coo_matrix((data, (row, col)), shape=(total_len, total_len))
+        return m, vertex, address_dict
+
+    def mat_all_point(self, m, vertex):
+        """
+            get E-alpha*m.T
+        """
+        total_len = len(vertex)
+        row = []
+        col = []
+        data = []
+        for index in range(total_len):
+            row.append(index)
+            col.append(index)
+            data.append(1)
+        row = np.array(row)
+        col = np.array(col)
+        data = np.array(data)
+        eye_t = coo_matrix((data, (row, col)), shape=(total_len, total_len))
+        return eye_t.tocsr() - self.alpha * m.tocsr().transpose()
+
+    def personal_rank_mat(self, personal_rank_graph, train_user_set):
+        m, vertex, address_dict = self.graph_to_mat(personal_rank_graph)
+        mat_all = self.mat_all_point(m, vertex)
+
+

utils/evaluate/user_recall_evalute.py

@@ -0,0 +1,6 @@
+# -*- encoding: utf-8 -*-
+"""
+@author:
+@date: 2022/10/13 14:14
+@brief:
+"""