前言
我在 Python 训练了个模型，怎么交给 Java 用呢？也介绍到了 m2cgen，今天碰巧看到了网上的一个把生成的 VBA 模型再转 SAS 代码的方案，现分享给大家。
m2cgen 是一个非常友好的包，可以将许多不同的训练模型转换为支持的语言[1]，如 R 和 VBA。但是，m2cgen 尚不支持 SAS。本文适用于需要在 SAS 环境中部署训练好的模型的人。本文介绍的赛道是先将模型转为 VBA 代码，再将 VBA 代码改成 SAS 脚本。
示例
将 XGBoost 模型转换为 VBA，然后转换为 SAS 脚本（有缺少值）
数据
从 sklearn 加载的 Iris 数据集
建模并转为 VBA，
# import packagesimport pandas as pdimport numpy as npimport osimport refrom sklearn import datasetsfrom xgboost import XGBClassifierfrom sklearn.model_selection import train_test_splitfrom sklearn.metrics import accuracy_scoreimport m2cgen as m2c# import datairis = datasets.load_iris()X = iris.dataY = iris.target# split data into train and test setsseed = 2020test_size = 0.3X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=test_size, random_state=seed)# fit model on training datamodel = XGBClassifier()model.fit(X_train, y_train)code = m2c.export_to_visual_basic(model, function_name = 'pred')
VBA 转 SAS
import re# remove unnecessary thingscode = re.sub('Dim var.* As Double', '', code)code = re.sub('End If', '', code)# change the script to sas scripts# change the beginningcode = re.sub('Module ModelFunction pred\(ByRef inputVector\(\) As Double\) As Double\(\)', 'DATA pred_result;SET dataset_name;', code)# change the endingcode = re.sub('End FunctionEnd Module', 'RUN;', code)# insert ';'all_match_list = re.findall('[0-9]+', code)for idx in range(len(all_match_list)): original_str = all_match_list[idx] new_str = all_match_list[idx][:-1]+';' code = code.replace(original_str, new_str)all_match_list = re.findall('\)', code)for idx in range(len(all_match_list)): original_str = all_match_list[idx] new_str = all_match_list[idx][:-1]+';' code = code.replace(original_str, new_str)# handle missing valuesall_match_list = re.findall('If.*Then', code)for idx in range(len(all_match_list)): original_str = all_match_list[idx] new_str = ' '.join(original_str.split()[:-1] + ['and not missing', original_str.split()[1], ' Then']) code = code.replace(original_str, new_str)# replace the 'inputVector' with var namedictionary = {'inputVector(0)':'sepal_length', 'inputVector(1)':'sepal_width', 'inputVector(2)':'petal_length', 'inputVector(3)':'petal_width'}for key in dictionary.keys(): code = code.replace(key, dictionary[key])# change the prediction labelscode = re.sub('Math.Exp', 'Exp', code)code = re.sub('pred = .*', '', code)temp_var_list = re.findall(r"var[0-9]+\(\d\)", code)for var_idx in range(len(temp_var_list)): code = re.sub(re.sub('\\(', '\\(', re.sub('\\)', '\\)', temp_var_list[var_idx])), iris.target_names[var_idx]+'_prob', code)# save outputwith open('vb1.sas', 'w') as vb vb.write(code)
生成的结果如下
详细代码可以点击 阅读原文。
参考资料
[1]
支持的语言: https://github.com/BayesWitnesses/m2cgen#supported-languages