This repository has been archived by the owner on Feb 6, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 29
/
Copy pathtrain_5_cross.py
105 lines (95 loc) · 6.14 KB
/
train_5_cross.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
import lightgbm as lgb # 模型
import pandas as pd # 数据处理包
import numpy as np # 数据处理包
from sklearn import metrics # 混淆句子
from sklearn.model_selection import StratifiedKFold, GridSearchCV, train_test_split # 分层五折验证包、寻找最优参函数、切分数据
from sklearn.metrics import accuracy_score, roc_curve, auc, confusion_matrix # 准确率、roc计算、auc计算、混淆矩阵
import itertools # 处理混淆矩阵
import gc # 处理缓存，有兴趣的可以搜搜怎么使用
import warnings # 忽略普通警告，不打印太多东西
warnings.filterwarnings('ignore')
def train_5_cross(df_pre, X,y, X_test_v1,y_test_v1, thresholds=0.45, id_1='id', csv_name=0):
"""
功能: 五折训练并输出名单
why: 5折一般是效果比较稳定的，用于线下做的。
X: 训练数据X（无标签/df型）
y: 训练数据y（标签/df型）
X_test_v1: 预测数据X（无标签/df型）
y_test_v1: 预测数据y（无标签/df型）
thresholds: 阈值选择，默认0.45高精确率
csv_name: 保存csv的名称，默认不保存
returen:
模型，客户名单及情况
"""
vali_auc_num=0 # 验证集AUC
vali_recall_num=0 # 验证集召回率
vali_precision_num=0 # 验证集精确率
test_auc_num=0 # 预测集AUC
test_recall_num=0 # 预测集召回率
test_precision_num=0 # 预测集精确率
y_pred_input = np.zeros(len(X_test_v1)) # 相应大小的零矩阵
print("=============开始训练================")
folds = StratifiedKFold(n_splits=5, shuffle=True, random_state=1234) # 分层采样, n_splits为几折
for fold_, (trn_idx, val_idx) in enumerate(folds.split(X, y)):
print("第 {} 次训练...".format(fold_+1))
train_x, trai_y = X.loc[trn_idx], y.loc[trn_idx]
vali_x, vali_y = X.loc[val_idx], y.loc[val_idx]
# 以下为调过参的lgb模型
clf = lgb.LGBMClassifier(max_depth=20, min_data_in_bin=5, max_bin=200,
min_child_samples=90, num_leaves=20, n_estimators=20000,
objective='binary', boosting_type='gbdt', learning_rate=0.02,
lambda_l2=5)
clf.fit(train_x, trai_y, eval_set=[(train_x, trai_y), (vali_x, vali_y)], verbose=0,
early_stopping_rounds=100, eval_metric='f1')
# 不懂的去GitHub看搜LightGBM的参数解释
# ===============验证集AUC操作===================
y_prb = clf.predict_proba(vali_x)[:,1] # 获取预测概率
# fpr:在实际为正的样本中，被正确判断为正的比例。tpr:在实际为负的样本中，被正确判断为负的比例。thres为阈值
fpr, tpr, thres = roc_curve(vali_y, y_prb)
vali_roc_auc = auc(fpr, tpr) # 获取验证集auc
vali_auc_num += vali_roc_auc # 将本次auc加入总值里
print("vali auc = {0:.4}".format(vali_roc_auc)) # 本次auc的值
# ===============预测集AUC操作===================
y_prb_test = clf.predict_proba(X_test_v1)[:,1] # 获取预测概率
fpr, tpr, thres = roc_curve(y_test_v1, y_prb_test)
test_roc_auc = auc(fpr, tpr)
test_auc_num += test_roc_auc
print("test auc = {0:.4}".format(test_roc_auc))
# ===============验证metric操作===================
y_pre_proba = clf.predict_proba(vali_x.values)
y_predictions = y_pre_proba[:, 1]>thresholds # 取阈值多少以上的为True
cnf_matrix = confusion_matrix(vali_y, y_predictions) # 建立矩阵
np.set_printoptions(precision=2) # 控制在两位数
vali_recall = '{0:.3f}'.format(cnf_matrix[1,1]/(cnf_matrix[1,0]+cnf_matrix[1,1])) # 召回率
vali_precision = '{0:.3f}'.format(cnf_matrix[1,1]/(cnf_matrix[0,1]+cnf_matrix[1,1])) # 精确率
print("vali_metric: ", vali_recall, vali_precision)
vali_recall_num += float(vali_recall) # 将本次召回率加入总值里
vali_precision_num += float(vali_precision) # 将本次精确率加入总值里
# ===============预测metric操作===================
y_pre_proba_test = clf.predict_proba(X_test_v1.values)
y_predictions_test = y_pre_proba_test[:, 1]>thresholds # 取阈值多少以上的为True
cnf_matrix_test = confusion_matrix(y_test_v1, y_predictions_test) # 建立矩阵
np.set_printoptions(precision=2) # 控制在两位数
test_recall = '{0:.3f}'.format(cnf_matrix_test[1,1]/(cnf_matrix_test[1,0]+cnf_matrix_test[1,1])) # 召回率
test_precision = '{0:.3f}'.format(cnf_matrix_test[1,1]/(cnf_matrix_test[0,1]+cnf_matrix_test[1,1])) # 精确率
print("test_metric: ", test_recall, test_precision)
test_recall_num += float(test_recall) # 将本次召回率加入总值里
test_precision_num += float(test_precision) # 将本次精确率加入总值里
y_pred_input += y_pre_proba_test[:, 1] # 将每次的预测的结果写入数组中
print("5折泛化，验证集AUC：{0:.3f}".format(vali_auc_num/5)) # 前面是做了5次相加，所以这次要除以5
print("5折泛化，预测集AUC：{0:.3f}".format(test_auc_num/5))
print("5折泛化，验证集recall：{0:.3f}".format(vali_recall_num/5))
print("5折泛化，验证集precision：{0:.3f}".format(vali_recall_num/5))
print("5折泛化，预测集recall：{0:.3f}".format(test_recall_num/5))
print("5折泛化，预测集precision：{0:.3f}".format(test_recall_num/5))
print("================开始输出名单==================")
y_pred_input_end = y_pred_input / 5 # 前面是做了5次相加，所以这次要除以5
y_pred_input_precision = y_pred_input_end > thresholds # 获取高精确率的标签
submission = pd.DataFrame({"id": df_pre[id_1],
"概率": y_pred_input_end,
"高精确": y_pred_input_precision})
if csv_name != 0:
submission.to_csv("%s预测名单.csv" % csv_name, index=False) # 保存
print("================输出名单名单==================")
print(submission.head(5))
return clf