English | 简体中文
AutoX支持上线部署的服务
- 当前AutoX支持二分类场景
- 支持回归及时序预测的功能正在开发中
autox_server在结构上分训练和预测两部分。
Step 1. 训练:通过对训练集进行数据探索,获得AutoX的解决方案(包含数据预处理、拼表、特征工程、模型选择、
模型调参、模型融合等),将解决方案以pickle各保存到指定路径中。
from autox import AutoXServer
# 指定配置文件路径
data_info_path = './bank/data_info.json'
# 指定训练数据路径
train_set_path = './bank/train_data'
# 初始化AutoXServer并执行AutoXServer训练模块, server_name自定义
autoxserver = AutoXServer(is_train = True, server_name = 'bank',
data_info_path = data_info_path, train_set_path = train_set_path)
autoxserver.fit()
# 指定路径并保存
autoxserver.save_server(save_path = './save_path')
Step 2. 预测:将导入训练好的AutoXServer,调用预测函数,获得预测结果。
from autox import AutoXServer
# 导入训练好的AutoXServer,注意server_name需要和训练时一致
autoxserver = AutoXServer(is_train = False, server_name = 'bank')
autoxserver.load_server(save_path = './save_path')
# 指定测试集路径,调用autoxserver的predict函数,返回预测结果(pandas dataframe格式)
pred = autoxserver.predict(test_set_path = ./bank/test_data)
案例描述:通过用户基本信息,消费行为,还款情况等,建立准确的逾期预测模型,以预测用户是否会逾期还款。
数据下载地址:百度网盘-提取码:phgb, google cloud
详细数据说明:link
autox_server训练代码:bank_train.ipynb
autox_server预测代码:bank_test.ipynb
一个符合要求的数据格式例子如下, 包括数据信息文件data_info.json、训练集文件夹、测试集文件夹:
data_info.json
train_data/
browse_train.csv
bill_train.csv
userinfo_train.csv
bank_train.csv
overdue_train.csv
test_data/
browse_test.csv
bill_test.csv
userinfo_test.csv
bank_test.csv
overdue_test.csv
- data_info.json包含数据基础信息、各表信息和表关系。
{
"dataset_id": "Athena", # 数据表名称
"recom_metrics": ["auc"], # 推荐的评价指标
"target_entity": "overdue", #主表(label列所在表为主表)
"target_id": ["new_user_id"], # id列
"target_label": "label", # 目标值列
"time_budget": 1200, # 时间预算
"entities": { # 数据集中各表的信息
"overdue": { # overdue表
"file_name": "overdue.csv", # 表名
"format": "csv", # 表格式
"header": "true", # 是否有header
"is_static": "true", # 是否是静态表
"time_col": [], # 时间列对应的列名,只有非静态表才有
"columns": # 列的数据类型
[{"new_user_id": "Str"},
{"label": "Num"},
{"flag1": "Num"},
{"mock_time": "Timestamp"},
{"mock_labelEncoder": "Str"}]
},
"userinfo": {
"file_name": "userinfo.csv",
"format": "csv",
"header": "true",
"is_static": "true",
"time_col": [],
"columns":
[{"new_user_id": "Str"}, {"flag1": "Num"}, {"flag2": "Num"}, {"flag3": "Num"}, {"flag4": "Num"}, {"flag5": "Num"}]
},
"bank": {
"file_name": "bank.csv",
"format": "csv",
"header": "true",
"is_static": "false",
"time_col": ["flag1"],
"columns":
[{"new_user_id": "Str"}, {"flag1": "Num"}, {"flag2": "Num"}, {"flag3": "Num"}, {"flag4": "Num"}]
},
"browse": {
"file_name": "browse.csv",
"format": "csv",
"header": "true",
"is_static": "false",
"time_col": ["flag1"],
"columns":
[{"new_user_id": "Str"}, {"flag1": "Num"}, {"flag2": "Num"}, {"flag3": "Num"}]
},
"bill": {
"file_name": "bill.csv",
"format": "csv",
"header": "true",
"is_static": "false",
"time_col": ["flag1"],
"columns":
[{"new_user_id": "Str"}, {"flag1": "Num"}, {"flag2": "Num"}, {"flag3": "Num"}, {"flag4": "Num"}, {"flag5": "Num"}, {"flag6": "Num"}, {"flag7": "Num"}, {"flag8": "Num"}, {"flag9": "Num"}, {"flag10": "Num"}, {"flag11": "Num"}, {"flag12": "Num"}, {"flag13": "Num"}, {"flag14": "Num"}]
}
},
"relations": [ # 表关系(可以包含为1-1, 1-M, M-1, M-M四种)
{
"related_to_main_table": "true", # 是否为和主表的关系
"left_entity": "overdue", # 左表名字
"left_on": ["new_user_id"], # 左表拼表键
"right_entity": "userinfo", # 右表名字
"right_on": ["new_user_id"], # 右表拼表键
"type": "1-1" # 左表与右表的连接关系
},
{
"related_to_main_table": "true",
"left_entity": "overdue",
"left_on": ["new_user_id"],
"left_time_col": "flag1",
"right_entity": "bank",
"right_on": ["new_user_id"],
"right_time_col": "flag1",
"type": "1-M"
},
{
"related_to_main_table": "true",
"left_entity": "overdue",
"left_on": ["new_user_id"],
"left_time_col": "flag1",
"right_entity": "browse",
"right_on": ["new_user_id"],
"right_time_col": "flag1",
"type": "1-M"
},
{
"related_to_main_table": "true",
"left_entity": "overdue",
"left_on": ["new_user_id"],
"left_time_col": "flag1",
"right_entity": "bill",
"right_on": ["new_user_id"],
"right_time_col": "flag1",
"type": "1-M"
}
]
}