DynamicRecSys/Tester.py at master · LightingTom/DynamicRecSys · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
import pickle


class Tester:

    def __init__(self, path):
        # Top K
        self.k = [5, 10, 20]

        self.session_length = 20 - 1
        self.pickle_path = path
        self.min_time = 1.0
        self.i_count = [0] * self.session_length
        self.recall = [[0] * len(self.k) for _ in range(self.session_length)]
        self.mrr = [[0] * len(self.k) for _ in range(self.session_length)]

        # 特殊对待第一项预测（因为第一项是根据inter_rnn给出的hidden state作为输入）
        self.first_recall = [0] * len(self.k)
        self.first_mrr = [0] * len(self.k)
        self.first_count = 0

        # 用于记录不同的范围的time gap结果的精确度
        self.time_buckets = [self.min_time, 2, 12, 36, 60, 84, 108, 132, 156, 180, 204, 228, 252, 276, 300, 348, 396,
                             444, 500, 501]
        self.time_count = [0] * len(self.time_buckets)
        self.mae = [0] * len(self.time_buckets)
        self.time_percent_mae = [0] * len(self.time_buckets)

    def reinitialize(self):
        self.i_count = [0] * self.session_length
        self.recall = [[0] * len(self.k) for _ in range(self.session_length)]
        self.mrr = [[0] * len(self.k) for _ in range(self.session_length)]
        self.first_recall = [0] * len(self.k)
        self.first_mrr = [0] * len(self.k)
        self.first_count = 0
        self.time_buckets = [self.min_time, 2, 12, 36, 60, 84, 108, 132, 156, 180, 204, 228, 252, 276, 300, 348, 396,
                             444, 500, 501]
        self.time_count = [0] * len(self.time_buckets)
        self.mae = [0] * len(self.time_buckets)
        self.time_percent_mae = [0] * len(self.time_buckets)

    @staticmethod
    def get_rank(target, prediction):
        for i in range(len(prediction)):
            if target == prediction[i]:
                return i + 1
        return -1

    def evaluate_sequence(self, predicted_sequence, target_sequence, seq_len):
        for i in range(seq_len):
            target = target_sequence[i]
            prediction = predicted_sequence[i]
            for j in range(len(self.k)):
                if target in prediction.data[:self.k[j]]:
                    self.recall[i][j] += 1
                    self.mrr[i][j] += 1.0 / self.get_rank(target, prediction.data[:self.k[j]])
            self.i_count[i] += 1

    def evaluate_first_item(self, predictions, target):
        for i in range(len(self.k)):
            if target in predictions.data[: self.k[i]]:
                self.first_recall[i] += 1
                self.first_mrr[i] += 1.0 / self.get_rank(target, predictions.data[: self.k[i]])
        self.first_count += 1

    def evaluate_batch(self, predictions, targets, seq_len, first_predictions, first_targets):
        for index in range(predictions):
            self.evaluate_sequence(predictions[index], targets[index], seq_len[index])
            self.evaluate_first_item(first_predictions[index], first_targets[index])

    def evaluate_time(self, time_prediction, time_target):
        for i in range(len(self.time_buckets)):
            if time_target < self.time_buckets[i] or i == len(self.time_buckets) - 1:
                self.time_count[i] += 1
                self.mae[i] += abs(time_prediction - time_target)
                # 忽略那些半小时以内的time gap(可近似的认为是同一session的操作)
                if time_target >= 0.5:
                    self.time_percent_mae[i] += 100 * (self.mae[i] / time_target)
                break

    def evaluate_batch_time(self, time_predictions, time_targets):
        for i in range(len(time_targets)):
            self.evaluate_time(time_predictions.data[i].item(), time_targets[i])

    def get_recall_mrr_result(self):
        res = "Cumulative\n"
        res += "Recall@5\tRecall@10\tRecall@20\tMRR@5\tMRR@10\tMRR@20\n"

        # 特殊对待每个session的第一个预测
        res += "First Item\n"
        recall_info = ""
        mrr_info = ""
        for i in range(len(self.k)):
            recall_num = self.first_recall[i] / self.first_count
            mrr_num = self.first_mrr[i] / self.first_count
            recall_info += str(round(recall_num, 4)) + '\t'
            mrr_info += str(round(mrr_num, 4)) + '\t'
        res += recall_info + mrr_info + '\n'

        recall = [0] * len(self.k)
        mrr = [0] * len(self.k)
        count = 0
        for i in range(self.session_length):
            recall_info = ""
            mrr_info = ""
            res += "i <= " + str(i + 1) + "\n"
            # 考虑前面所有i个位置的效果，取其平均
            count += self.i_count[i]
            for j in range(len(self.k)):
                recall[j] += self.recall[i][j]
                mrr[j] += self.mrr[i][j]
                recall_info += str(round(recall[j] / count, 4)) + "\t"
                mrr_info += str(round(mrr[j] / count, 4)) + "\t"
            res += recall_info + mrr_info + "\n"
        return res

    def get_individual_recall_mrr_result(self):
        res = "Individual\n"
        res += "Recall@5\tRecall@10\tRecall@20\tMRR@5\tMRR@10\tMRR@20\n"
        for i in range(self.session_length):
            recall_info = ""
            mrr_info = ""
            res += "i <= " + str(i + 1) + "\n"
            for j in range(len(self.k)):
                recall_info += str(round(self.recall[i][j] / self.i_count[i], 4)) + "\t"
                mrr_info += str(round(self.mrr[i][j] / self.i_count[i], 4)) + "\t"
            res += recall_info + mrr_info + "\n"
        return res

    def get_time_result(self):
        res = "MAE\tPercent"
        total_count = 0
        total_mae = 0
        total_percent_mae = 0

        for i in range(len(self.time_buckets)):
            total_count += max(self.time_count[i], 1)
            total_mae += self.mae[i]
            total_percent_mae += self.time_percent_mae[i]

            res += "days<=" + self.time_buckets[i] + "\n"
            res += str(round(self.mae[i] / max(self.time_count[i], 1), 4)) + "\t"
            res += str(round(self.time_percent_mae[i] / max(self.time_count[i], 1), 4)) + "\n"
        res += "Total:\n"
        res += str(round(total_mae / max(total_count, 1), 4)) + "\t"
        res += str(round(total_percent_mae / max(total_count, 1), 4)) + "\n"
        return res

    def get_result(self, if_get_time=True):
        cumulate_recall_mrr_res = self.get_recall_mrr_result()
        individual_recall_mrr_res = self.get_individual_recall_mrr_result()
        time_res = ""
        if if_get_time:
            time_res = self.get_time_result()
        return cumulate_recall_mrr_res, time_res, individual_recall_mrr_res

    def store_result(self):
        mrr_recall_res = {"i_count": self.i_count, "k": self.k, "session_length": self.session_length, "mrr": self.mrr,
                          "recall": self.recall, "first_count": self.first_count, "first_mrr": self.first_mrr,
                          "first_recall": self.first_recall}
        time_res = {"mae": self.mae, "count": self.time_count, "time_buckets": self.time_buckets,
                    "time_percent_mae": self.time_percent_mae}
        res = {"mrr_recall_res": mrr_recall_res, "time_res": time_res}
        pickle.dump(res, open(self.pickle_path + ".pickle", "wb"))

    def get_result_and_reset(self, store=True):
        res = self.get_result()
        if store:
            self.store_result()
        self.reinitialize()
        return res