使用ubuntu python 拟合PMML文件格式模型

用案例解释如何生成PMML模型文件。
在ubuntu python下，拟合XGBoost模型。并在python 中添加 JAVA SDK 环境路径，导出支持JAVA的PMML模型文件，以便用于将模型部署在JAVA 平台。

1.安装ubuntu 版本 jdk

1.1从官网中下载对应的jdk版本
https://download.oracle.com/otn/java/jdk/8u361-b09/0ae14417abb444ebb02b9815e2103550/jdk-8u361-linux-x64.tar.gz?AuthParam=1680089271_bcf63bda2c61968fb00eda136bdea84a

1.2 指定安装包存放地址，并解压

/jvm# cp  /quant/联合建模/jdk-8u361-linux-x64.tar.gz   java.tar.gz
root@hecs-87892:/usr/lib/jvm# sudo tar -C /usr/lib/jvm -xzf java.tar.gz
root@hecs-87892:/usr/lib/jvm# cd /usr/lib/jvm
root@hecs-87892:/usr/lib/jvm# ls

1.3 激活jdk 配置环境变量

cd /etc
vim profile
#打开profile 后并插入
export JAVA_HOME=/usr/lib/jvm/jdk1.8.0_361
export JRE_HOME=${JAVA_HOME}/jre
export CLASSPATH=.:${JAVA_HOME}/lib:${JRE_HOME}/lib
export PATH=${JAVA_HOME}/bin:$PATH
#完成插入，并:wq保存
source profile 
#检查java 是否安装

2.拟合PMML 格式模型

"""
part1:直接用PMML拟合
ubuntu 10.04.4 LTS  X86_64
python==3.9
sklearn2pmml==0.91.1
scikit-learn==1.2.2
xgboost==0.90
joblib==1.2.0
numpy==1.20.3
"""



import numpy as np
from sklearn2pmml import *
import joblib

import pandas as pd 
from xgboost  import XGBClassifier

import os
os.environ['PATH']+=os.pathsep + '/usr/lib/jvm/jdk1.8.0_361/bin' # windows 用小写 path

def to_score(x):
    import math
    if x <=0.001:
        x =0.001
    elif x >=0.999:
        x =0.999
        
    A = 598.930226033773
    B = 43.280851226669
    #PD0=25
    result = round(A-B*math.log(x/(1-x)),2)
    
    if result < 300:
        result=300
    if result>900:
        result=900
    return result


params={'objective': 'binary:logistic',
 'base_score': 0.5,
 'booster': 'gbtree',
 'colsample_bylevel': 1,
 'colsample_bynode': 1,
 'colsample_bytree': 0.5,
 'gamma': 0.0,
 'importance_type': 'gain',
 'learning_rate': 0.1,
 'max_delta_step': 0,
 'max_depth': 1,
 'min_child_weight': 88,
 'missing': np.nan,
 'n_estimators': 104,
 'random_state': 0,
 'reg_alpha': 1e-05,
 'reg_lambda': 1e-05,
 'scale_pos_weight': 1,
 'subsample': 0.5,
 'num_leaves': 2}




df = pd.read_csv('第二批授信样本模型跑分结果.csv', encoding='utf-8-sig') 
label=pd.read_csv('y_label.csv', encoding='gb2312') 

label.columns
label=label.drop_duplicates(subset=('sm3_id'))

df.columns

all_df=df.merge(label[['label','sm3_id']],how='inner',left_on='certif_id_sm3',right_on='sm3_id')
feat=list(all_df.columns[:88])

df[feat]=df[feat].replace(-1,np.nan)  

train=all_df[all_df['type']=='train']
X_train=train[feat]
y_train=train['label']


bst=XGBClassifier(**params)
pipeline=PMMLPipeline([('classifier',bst)])

pipeline.fit(X_train,y_train)

proba=pipeline.predict_proba(all_df[feat])[:,1]

score=[ to_score(i) for i in proba]
all_df['new_score']=score

sklearn2pmml(pipeline,r'ysd_sx_func_xgb_4_model.pmml',with_repr=True)  #需要安装