调参

在利用gridseachcv进行调参时，其中关于scoring可以填的参数在SKlearn中没有写清楚，就自己找了下，具体如下：

parameters = {'eps':[0.3,0.4,0.5,0.6], 'min_samples':[20,30,40]}
db = DBSCAN(metric='cosine', algorithm='brute').fit(xx)
grid = GridSearchCV(db, parameters, cv=5, scoring='adjusted_rand_score')

Scoring	Function	Comment
Classification
‘accuracy’	`metrics.accuracy_score`
‘average_precision’	`metrics.average_precision_score`
‘f1’	`metrics.f1_score`	for binary targets
‘f1_micro’	`metrics.f1_score`	micro-averaged
‘f1_macro’	`metrics.f1_score`	macro-averaged
‘f1_weighted’	`metrics.f1_score`	weighted average
‘f1_samples’	`metrics.f1_score`	by multilabel sample
‘neg_log_loss’	`metrics.log_loss`	requires `predict_proba` support
‘precision’ etc.	`metrics.precision_score`	suffixes apply as with ‘f1’
‘recall’ etc.	`metrics.recall_score`	suffixes apply as with ‘f1’
‘roc_auc’	`metrics.roc_auc_score`
Clustering
‘adjusted_rand_score’	`metrics.adjusted_rand_score`
Regression
‘neg_mean_absolute_error’	`metrics.mean_absolute_error`
‘neg_mean_squared_error’	`metrics.mean_squared_error`
‘neg_median_absolute_error’	`metrics.median_absolute_error`
‘r2’	`metrics.r2_score`

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

但后面听另外一个课的时候老师说，对于特征较多的模型不建议用gridSearch ，耗时，而且只是在train上表现好的参数，不一定在跨时间验证集上表现好

建议设计调参，设计的目标是跨时间验证集的KS要最大化，同时跨时间验证集和训练集的KS差距最小

调参方法

offks + 0.8(offks - devks)最大化

import pandas as pd
from sklearn.metrics import roc_auc_score,roc_curve,auc
from sklearn.model_selection import train_test_split
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
import numpy as np
import random
import math
import lightgbm as lgb
from sklearn.model_selection import train_test_split

data = pd.read_csv('Acard.txt')

train = data[data.obs_mth != '2018-11-30'].reset_index().copy()
val = data[data.obs_mth == '2018-11-30'].reset_index().copy()
feature_lst = ['person_info','finance_info','credit_info','act_info']
x = train[feature_lst]
y = train['bad_ind']

val_x =  val[feature_lst]
val_y = val['bad_ind']


train_x,test_x,train_y,test_y = train_test_split(x,y,random_state=0,test_size=0.2)

#改变我们想去调整的参数为value，设置调参区间
min_value = 40
max_value = 60
for value in  range(min_value,max_value+1):
    best_omd = -1
    best_value = -1
    best_ks=[]
    def  lgb_test(train_x,train_y,test_x,test_y):
        clf =lgb.LGBMClassifier(boosting_type = 'gbdt',
                               objective = 'binary',
                               metric = 'auc',
                               learning_rate = 0.1,
                               n_estimators = value,
                               max_depth = 5,
                               num_leaves = 20,
                               max_bin = 45,
                               min_data_in_leaf = 6,
                               bagging_fraction = 0.6,
                               bagging_freq = 0,
                               feature_fraction = 0.8,
                               silent=True
                               )
        clf.fit(train_x,train_y,eval_set = [(train_x,train_y),(test_x,test_y)],eval_metric = 'auc')
        return clf,clf.best_score_['valid_1']['auc'],
    lgb_model , lgb_auc  = lgb_test(train_x,train_y,test_x,test_y)

    y_pred = lgb_model.predict_proba(x)[:,1]
    fpr_lgb_train,tpr_lgb_train,_ = roc_curve(y,y_pred)
    train_ks = abs(fpr_lgb_train - tpr_lgb_train).max()

    y_pred = lgb_model.predict_proba(val_x)[:,1]
    fpr_lgb,tpr_lgb,_ = roc_curve(val_y,y_pred)
    val_ks = abs(fpr_lgb - tpr_lgb).max()
    
    Omd= val_ks + 0.8*(val_ks - train_ks)
    if Omd>best_omd:
        best_omd = Omd
        best_value = value
        best_ks = [train_ks,val_ks]
print('best_value:',best_value)
print('best_ks:',best_ks)

调参

推荐阅读