PIER/config.py at main · dwLyong/PIER · 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
# coding: utf-8 -*-
import shutil
import time
import os

# result
RANDOM_SEED = 2022
BATCH_SIZE = 1024
IMP_LOSS_WEIGHT = 0.02
# basic config
EPOCH = 1
LEARNING_RATE = 0.001
DATA_MODE = 3 # 1:local train，2:local test, 3:docker evaluate
TRAIN_MODE = 3
MODEL_NAME = "pier_listwise_whole_v4"

USE_CONSTRATIVE_LOSS = True
CONSTRATIVE_LOSS_K = 0.01

# poi类别特征
FEATURE_CATE_NUM = 7 # v1r3:19
# dense特征
FEATURE_DENSE_NUM = 5  # v1:28 v1r2:79 v1r3:83
# 预估值特征
FEATURE_CXR_NUM = 3
# 环境特征
FEATURE_ENV_NUM = 2
# 自然poi
FEATURE_NATURE_POI = 25

# embedding_look_up维度
CATE_FEATURE_EMBEDDINGS_SHAPE = [1 << 22, 8]

# hash vector
HASH_VECTOR = [[0.4011729380078948, 0.9907933488085712, 0.889170658914898, 0.015553701343792192, 0.2269008585804, 0.889609750280199, 0.19280345796299014, 0.33362195188257815],
               [0.1174580997476552, 0.4475570392889796, 0.9925939893135071, 0.7296640075216434, 0.6436313332140967, 0.8568068597644793, 0.9018604021483339, 0.28774343258170776],
               [0.4374993384759095, 0.6807149381274915, 0.20502126763268802, 0.6968687323602859, 0.6449029002244834, 0.08732080642447282, 0.6119476780855001, 0.8616788453789646],
               [0.19344266090314144, 0.4268233179146762, 0.10951597767480326, 0.9867838283258178, 0.8340011944969644, 0.7992329879482085, 0.005303560724105649, 0.9662924610057512],
               [0.5824282763301396, 0.5090708710943849, 0.4462703076663568, 0.5482943153972023, 0.06782829736851825, 0.08907408658464577, 0.13400486343251583, 0.1848228429919272]]

POSITION_ENCODING = [[0.08790239841717873, 0.033267486152506076, 0.5495130189114207, 0.17419777583517537, 0.6838981992197484, 0.07935154925635501, 0.02665372302227631, 0.4181008411574786],
                     [0.02038159041970189, 0.9310485215006198, 0.723661313625571, 0.5110323516977285, 0.7812851438212606, 0.40722710713947474, 0.04646490014890503, 0.9565353323381218],
                     [0.17279361698259843, 0.12968275664201512, 0.3183103529758954, 0.6360591081256931, 0.1558507653689548, 0.5972802646455662, 0.4380619835390329, 0.6088094249662641],
                     [0.038262298606345335, 0.9999545626217287, 0.7113776275017341, 0.5434618368150265, 0.6853527957705402, 0.9662653254145415, 0.9641592716989676, 0.3443813983264],
                     [0.7903243938847678, 0.9952713339078417, 0.8741415264071601, 0.45665348276461737, 0.7693872696125916, 0.772509599868299, 0.2540369924156157, 0.24781240400239857]]


TIME_AWARE_WEIGHT = [[1/2,1/2,1/3,1/3,1/4]]
PERMUATION_SIZE = 120
TOP_K = 20


EXPOSE_RATE_FOR_BEAM_SEARCH = [[1.0,0.9,0.8,0.7,0.6]]

# N: Cut Number of POI For Train
POI_NUM = 5
FEATURE_NUM = 9
PAGE_NUM = 5
FEATURE_NUM_FOR_PAGE = 11

# 属性特征：KA AOR BRAND
FEATURE_ATTR_NUM = 3

# DELIVERY_FEAT
DELIVERY_FEAT_NUM = 4

# OUT NUM
OUT_NUM = 1

PLACE_HOLDER_NUM = 11
DENSE_FEAT_NUM = 439


# 网络结构参数
MODEL_PARAMS = {
    'INPUT_TENSOR_LAYERS_A': [60, 32, 20],
    'INPUT_TENSOR_LAYERS_B': [128, 32],
    'INPUT_TENSOR_LAYERS_C': [50, 20],
    'INPUT_TENSOR_LAYERS_D': [50, 20],
    'INPUT_TENSOR_LAYERS_E': [50, 20]
}
# A_INPUT_DIM = POI_NUM * (MODEL_PARAMS['INPUT_TENSOR_LAYERS_A'][-1] + 1)
MLP_INPUT_DIM = CATE_FEATURE_EMBEDDINGS_SHAPE[1] * 3 + 1 + MODEL_PARAMS['INPUT_TENSOR_LAYERS_A'][-1]
DIN_CONF = {}

# train data
# # /users/lemonace/Downloads/tfrecord-rl-limit5-v1
# if DATA_MODE == 1:
#     TRAIN_FILE = ['/users/meituan_sxw/Downloads/part-r-00046']
#     VALID_FILE = TRAIN_FILE
#     PREDICT_FILE = VALID_FILE
#     TEST_FILE = PREDICT_FILE
# elif DATA_MODE == 2:
#     TRAIN_FILE = ['/mnt/dolphinfs/hdd_pool/docker/user/hadoop-hmart-waimaiad/yangfan129/train_data/avito_v1_new/avito_v1_new/test_data/part-r-*']
#     VALID_FILE = TRAIN_FILE
#     TEST_FILE = VALID_FILE
# elif DATA_MODE == 3:
#     TRAIN_FILE = ["/home/liushuchang/.jupyter/hailan_workspace/avito_v3_new/train_data/part-r-*"]
#     VALID_FILE = ["/home/liushuchang/.jupyter/hailan_workspace/avito_v3_new/test_data/part-r-*"]
#     TEST_FILE= ["/home/liushuchang/.jupyter/hailan_workspace/avito_v3_new/test_data/part-r-*"]
# elif DATA_MODE == 4:
#     DATA_FILE = "/home/hadoop-hmart-waimaiad/cephfs/data/yangfan129/train_data/tfrecord-multi-channel-v1/"
#     #TRAIN_LIST = ["20211222", "20211223", "20211224", "20211225"]
#     TRAIN_LIST = ["20220123"]
#     VALID_LIST = ["20220124"]
#     TRAIN_FILE = [DATA_FILE + x + "/part-r-*" for x in TRAIN_LIST]
#     VALID_FILE = [DATA_FILE + x + "/part-r-0001*" for x in VALID_LIST]
#     TEST_FILE = [DATA_FILE + x + "/part-r-00011" for x in TRAIN_LIST]
#
# # 辅助脚本
# MEAN_VAR_PATH_POI = "./avg_std/poi"
# MEAN_VAR_PATH_DELIVERY = "./avg_std/delivery"
# MODEL_SAVE_PATH = "./model/" + MODEL_NAME
# MODEL_SAVE_PB_EPOCH_ON = False
# MODEL_SAVE_PB_EPOCH_PATH = MODEL_SAVE_PATH + "_pbs"

if DATA_MODE == 1:
    TRAIN_FILE = ['/users/meituan_sxw/Downloads/part-r-00046']
    VALID_FILE = TRAIN_FILE
    PREDICT_FILE = VALID_FILE
    TEST_FILE = PREDICT_FILE
elif DATA_MODE == 2:
    TRAIN_FILE = ['/mnt/dolphinfs/hdd_pool/docker/user/hadoop-hmart-waimaiad/yangfan129/train_data/avito_v1_new/avito_v1_new/test_data/part-r-*']
    VALID_FILE = TRAIN_FILE
    TEST_FILE = VALID_FILE
elif DATA_MODE == 3:
    TRAIN_FILE = ["/home/liushuchang/.jupyter/hailan_workspace/avito_v3_new/train_data/part-r-*"]
    VALID_FILE = ["/home/liushuchang/.jupyter/hailan_workspace/avito_v3_new/test_data/part-r-*"]
    TEST_FILE= ["/home/liushuchang/.jupyter/hailan_workspace/avito_v3_new/test_data/part-r-*"]
elif DATA_MODE == 4:
    DATA_FILE = "/home/hadoop-hmart-waimaiad/cephfs/data/yangfan129/train_data/tfrecord-multi-channel-v1/"
    #TRAIN_LIST = ["20211222", "20211223", "20211224", "20211225"]
    TRAIN_LIST = ["20220123"]
    VALID_LIST = ["20220124"]
    TRAIN_FILE = [DATA_FILE + x + "/part-r-*" for x in TRAIN_LIST]
    VALID_FILE = [DATA_FILE + x + "/part-r-0001*" for x in VALID_LIST]
    TEST_FILE = [DATA_FILE + x + "/part-r-00011" for x in TRAIN_LIST]

# 辅助脚本
MEAN_VAR_PATH_POI = "./avg_std/poi"
MEAN_VAR_PATH_DELIVERY = "./avg_std/delivery"
MODEL_SAVE_PATH = "./model/" + MODEL_NAME
MODEL_SAVE_PB_EPOCH_ON = False
MODEL_SAVE_PB_EPOCH_PATH = MODEL_SAVE_PATH + "_pbs"