-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathtextdeepfm.py
156 lines (127 loc) · 6.46 KB
/
textdeepfm.py
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
"""
TextDeepFM模型
在DeepFM模型的基础上加入文本特征
将用户感兴趣的话题作为用户的embedding
将问题绑定的话题作为问题的embedding
"""
import pandas as pd
import numpy as np
from collections import OrderedDict, namedtuple
from sklearn.preprocessing import LabelEncoder, MinMaxScaler
from sklearn.model_selection import train_test_split
from itertools import chain
import torch
import time
import matplotlib.pyplot as plt
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data as Data
from sklearn.metrics import *
from torch.utils.data import DataLoader
from tqdm import tqdm
from sklearn.preprocessing import LabelEncoder
import math
import itertools
from utils import *
from gensim.models import KeyedVectors
import pickle
import gc
""" 运行TextDeepFM """
path = 'E:\\competition\\看山杯\\data\\data_set_0926\\'
train = pd.read_csv(path + 'train.txt', sep='\t')
test = pd.read_csv(path + 'test.txt', sep='\t')
data = pd.concat([train, test], ignore_index=True, sort=False)
# print(data[['m_num_interest_topic', 'q_num_topic_words', 'm_num_atten_topic']].describe()) maxlen 10 ,13, 100
"""单值类别特征、数值特征处理"""
# 单值类别特征
fixlen_category_columns = ['m_sex', 'm_access_frequencies', 'm_twoA', 'm_twoB', 'm_twoC', 'm_twoD', 'm_twoE', 'm_categoryA', 'm_categoryA',
'm_categoryB', 'm_categoryC', 'm_categoryD', 'm_categoryE', 'm_num_interest_topic', 'num_topic_attention_intersection',
'q_num_topic_words', 'num_topic_interest_intersection'
]
# 数值特征
fixlen_number_columns = ['m_salt_score', 'm_num_atten_topic', 'q_num_title_chars_words', 'q_num_desc_chars_words', 'q_num_desc_words', 'q_num_title_words',
'days_to_invite'
]
target = ['label']
data[fixlen_category_columns] = data[fixlen_category_columns].fillna('-1', )
data[fixlen_number_columns] = data[fixlen_number_columns].fillna(0, )
for feat in fixlen_category_columns:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
mms = MinMaxScaler(feature_range=(0, 1))
data[fixlen_number_columns] = mms.fit_transform(data[fixlen_number_columns])
"""文本特征处理"""
# 截断补齐
def pad_sequences(x, maxlen):
x = str(x)
if x == 'nan':
return [0] * maxlen
words = []
for t in x.split(" "):
words.append(int(t))
words = words[0:maxlen]
words += [0] * (maxlen-len(words))
return words
def deal_text(values, maxlen=10):
temp = []
for text in values:
temp.append(pad_sequences(text, maxlen))
return np.array(temp)
# 初始化词向量矩阵
topic2index = pickle.load(open(path+"topic2index.pick", 'rb'))
topic_word_vector = KeyedVectors.load_word2vec_format(path + "topic_vectors_64d.txt", binary=False)
embedding_matrix = np.random.uniform(size=(len(topic2index) + 1, 64))
miss_count = 0
for word, index in topic2index.items():
try:
word_vector = None
word_vector = topic_word_vector[word]
embedding_matrix[index] = word_vector
except:
miss_count += 1
print(miss_count, " 个词没有词向量")
text_feature_columns = [TextFeat('q_topic_words', len(topic2index) + 1, 10),
# TextFeat('m_attention_topics_words', len(topic2index) + 1, 10),
TextFeat('m_interested_topics_words', len(topic2index) + 1, 10)
]
fixlen_feature_columns = [SparseFeat(feat, data[feat].nunique())
for feat in fixlen_category_columns] + [DenseFeat(feat, 1,)for feat in fixlen_number_columns]
dnn_feature_columns = fixlen_feature_columns + text_feature_columns
linear_feature_columns = fixlen_feature_columns + text_feature_columns
feature_names = get_feature_names(linear_feature_columns + dnn_feature_columns)
train = data[~data['label'].isnull()]
test = data[data['label'].isnull()]
del train['time']
del train['memberID']
del train['questionID']
del train['m_attention_topics_words']
del test['m_attention_topics_words']
gc.collect()
train, vaild = train_test_split(train, test_size=0.2)
train_model_input = {name: train[name] for name in feature_names}
vaild_model_input = {name: vaild[name] for name in feature_names}
train_model_input['q_topic_words'] = deal_text(train_model_input['q_topic_words'])
# train_model_input['m_attention_topics_words'] = deal_text(train_model_input['m_attention_topics_words'])
train_model_input['m_interested_topics_words'] = deal_text(train_model_input['m_interested_topics_words'])
vaild_model_input['q_topic_words'] = deal_text(vaild_model_input['q_topic_words'])
# vaild_model_input['m_attention_topics_words'] = deal_text(vaild_model_input['m_attention_topics_words'])
vaild_model_input['m_interested_topics_words'] = deal_text(vaild_model_input['m_interested_topics_words'])
gc.collect()
device = 'cuda:0'
"""第一步：初始化一个模型类"""
model = TextDeepFM(linear_feature_columns=linear_feature_columns, dnn_feature_columns=dnn_feature_columns, task='binary',
l2_reg_embedding=1e-5, device=device, use_textcnn=True, filter_sizes=[1, 2, 3], num_filters=64, text_vocab_size=len(topic2index)+1, text_embedding_size=64, text_embedding_matrix=embedding_matrix, cnn_dropout=0.0)
"""第二步：调用compile()函数配置模型的优化器、损失函数、评价函数"""
model.compile("adam", "binary_crossentropy", metrics=["binary_crossentropy", "auc"],)
"""第三步：调用fit()函数训练模型"""
model.fit(train_model_input, train[target].values, batch_size=2048, epochs=40, validation_data=[vaild_model_input, vaild[target].values], verbose=1, model_cache_path='E:\\competition\\看山杯\\models\\textdeepfm.model', early_stop=2)
"""预测"""
test_model_input = {name: test[name] for name in feature_names}
test_model_input['q_topic_words'] = deal_text(test_model_input['q_topic_words'])
# test_model_input['m_attention_topics_words'] = deal_text(test_model_input['m_attention_topics_words'])
test_model_input['m_interested_topics_words'] = deal_text(test_model_input['m_interested_topics_words'])
pred_ans = model.predict(test_model_input, 2048)
pred_ans = pred_ans.reshape(pred_ans.shape[0])
result = test[['questionID', 'memberID', 'time']]
result['result'] = pred_ans
result.to_csv(path + 'submit.txt', sep='\t', index=False) # 注意提交的时候请把表头去掉