def __init__(self):
super().__init__()
self.queue_flow = mm.get_queue(cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivate = Mongo(**cm.config['mongo_jd_cl'])
self.rule_config_list = self.get_rule_list()
self.group_tag_config_list = self.get_group_tag_list()
class JDXJudger(MessageServer):
def __init__(self):
super().__init__()
self.queue_flow = mm.get_queue(cm.config['queue']['flow'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
def handle_msg(self, msg_dict):
app_id = msg_dict['app_id']
user_id = msg_dict['user_id']
category_id = msg_dict['category_id']
rule_result = msg_dict['rule_result'] #eg: 'MD1!MB9'
model_pass = msg_dict['model_pass']
server = msg_dict['server']
rule_pass = 0 if rule_result else 1
if rule_result == '' and model_pass == 1:
final_result = 1
result = 'Approved'
rejectReason = ''
elif rule_result != '' and model_pass == 1:
final_result = 0
result = 'Rejected'
rejectReason = rule_result
elif rule_result == '' and model_pass == 0:
final_result = 0
result = 'Rejected'
rejectReason = 'MODL'
else:
final_result = 0
result = 'Rejected'
rejectReason = rule_result + '!MODL'
self.save_judge_results(app_id, user_id, category_id, model_pass,
rule_pass, final_result, rejectReason, server)
msg = {
'app_id': app_id,
'job_name': msg_dict['job_name'],
'final_result': final_result,
'result': result,
"rejectReason": rejectReason
}
self.queue_flow.send_message(msg)
def save_judge_results(self,
app_id,
user_id,
category_id,
model_pass,
rule_pass,
final_pass,
reason,
server,
table='jdx_judge_results'):
sql = "insert into {} (app_id, user_id, category_id, model_pass, rule_pass, final_pass, reason, server) values " \
"('{}', '{}', '{}', {}, {}, {},'{}', '{}')".format(
table, app_id, user_id, category_id, model_pass, rule_pass, final_pass, reason, server)
self.mysql_ml.update(sql)
def __init__(self):
super(JDXMLWithdrawModelServer, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_der = Mongo(**cm.config['mongo_jd_cl'])
self.model_config_list = self.get_model_list()
self.group_tag_threshold = self.get_group_tag_threshold()
def __init__(self):
super(JDXMLCreditServer, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_der = Mongo(**cm.config['mongo_jd_cl'])
self.model_config_list = self.get_model_list()
self.target = 'jd'
def __init__(self):
super(JDXModifyQuota, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivable = Mongo(**cm.config['mongo_jd_cl'])
self.quota_modify_strategy_list = self.get_quota_modify_strategy_list(
) # 获取全部额度变更策略配置
def read_mysql(self, sql):
mysql_client = MySql(host='10.47.89.174',
port=3229,
user='******',
pw='gGynbdutOr3Eeg0I',
db='maple_leaf')
res = mysql_client.query(sql=sql)
return pd.DataFrame(res)
def __init__(self):
# db client
self.mysql_mlx_client = MySql(**cm.config['mysql_jd_cl'])
self.mongo_mlx_client = Mongo(**cm.config['mongo_jd_cl'])
# db access
self.ml_result_access = jdx_data.MachineLearningJdResults(
client=self.mysql_mlx_client)
self.original_data_access = jdx_data.DerivativeProdData(
client=self.mongo_mlx_client)
self.threshold_access = jdx_data.PrThreshold(
client=self.mysql_mlx_client)
def __init__(self, strategy, *args, **argw):
self.name = strategy.get('name')
self.model_path = strategy.get('model_path')
self.match = strategy.get('match')
self.params = strategy.get('params')
self.output = strategy.get('output')
self.prob = strategy.get('prob')
self.is_setup = strategy.get('is_setup')
self.priority = strategy.get('priority')
self.mysql_mlx_client = MySql(**cm.config['mysql_jd_cl'])
self.mongo_mlx_client = Mongo(**cm.config['mongo_jd_cl'])
self.original_data_access = jdx_data.DerivativeProdData(
client=self.mongo_mlx_client)
self.strategy_result_access = jdx_data.CreditAmountStragetyResult(
client=self.mysql_mlx_client)
self._trans_params()
class JDXMLWithdrawModelServer(message_server.MessageServer):
def __init__(self):
super(JDXMLWithdrawModelServer, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_der = Mongo(**cm.config['mongo_jd_cl'])
self.model_config_list = self.get_model_list()
self.group_tag_threshold = self.get_group_tag_threshold()
def handle_msg(self, msg_dict):
if 'app_id' not in msg_dict:
logging.warning('No key in msg called appId!')
app_id = str(msg_dict['app_id']).lower()
user_id = msg_dict['user_id']
category_id = msg_dict['category_id']
group_tag = msg_dict['group_tag']
model = self.choose_model(category_id, 'jd')
feature_data = self.get_der_data(app_id)
creditLine = float(feature_data.get(
'X_JD_CreditLine')) if feature_data.get('X_JD_CreditLine') else -1
principal = float(feature_data.get(
'X_JD_Principal')) if feature_data.get('X_JD_Principal') else -1
# 是否是预审核
is_preAudit = 1 if feature_data.get(
'X_WorkFlow') == 'buyApplicationPreAudit' or feature_data.get(
'X_isPreAudit') == 'preAudit' else 0
model_score = self.run_model(model, feature_data)
op_score = self.get_op_score(model_score, model['op_coef'],
model['op_intercept'])
model_pass = 0 if model_score < model['threshold'] else 1
# 根据op_score判别控制组
if group_tag == 'B' and model_pass == 0 and op_score >= self.group_tag_threshold:
model_pass = 1
self.save_model_result(app_id, user_id, category_id, creditLine,
principal, model_score, model['threshold'],
model['model_name'], is_preAudit)
self.save_op_socre_result(app_id, user_id, 'jd', 'withdraw',
category_id, model['model_name'],
model_score, op_score)
msg = {
"app_id": app_id,
"job_name": msg_dict['job_name'],
"model_pass": model_pass,
"op_score": op_score
}
self.queue_flow.send_message(msg)
def choose_model(self, category_id, target):
models = []
for model in self.model_config_list:
if model['category_id'] == category_id and model[
'target'] == target:
models.append(model)
rand = np.random.random()
model_prod = 0
for model in models:
model_prod += model['prob']
if rand < model_prod:
return model
return None
def get_model_list(self, table='jdx_category_model_relations'):
sql = """select category_id,model_name,target,estor_path,threshold,op_coef,op_intercept,prob from {} where is_setup=1;""".format(
table)
model_list = self.mysql_ml.query(sql)
return model_list
def get_group_tag_threshold(self, table='jdx_proba_tags'):
sql = """select threshold from {} where flow_type='w' and is_setup=1;""".format(
table)
group_tag_threshold = self.mysql_ml.query_one(sql)
return float(
group_tag_threshold['threshold']) if group_tag_threshold else None
def run_model(self, chosed_model, feature_data):
"""跑模型，获得模型分数"""
model_path = chosed_model['estor_path']
with open(model_path, 'rb') as f:
pk_obj = pickle.load(f)
model = pk_obj['model']
agg_attr = []
fields = model.get_params()['enum'].clean_col_names
for attr in fields:
agg_attr.append(feature_data.get(attr))
# use pandas to format data, the dimension should be 1 x len(fields)
result_df = pd.DataFrame(agg_attr, index=fields).T
assert result_df.ndim == 2, 'result_df should be two dimension'
# transform data and make prediction
pred_probas = model.predict_proba(result_df)
model_score = pred_probas[:, 1][0]
return float(model_score)
def get_der_data(self, app_id, collection='derivables'):
"""获取特征数据"""
result = {}
try:
result = self.mongo_der.get_collection(collection).find({
'_id':
app_id.upper()
}).next()
result = {key: value for key, value in result.items()}
except Exception:
pass
return result
def get_op_score(self, model_score, op_coef, op_intercept):
if not (op_coef and op_intercept):
return 0
model_ln_odds = math.log(model_score / (1 - model_score)) / math.log(2)
ln_real_odds = op_coef * model_ln_odds + op_intercept
op_score = max(min(60 * ln_real_odds + 300, 1000), 0)
return op_score
def save_model_result(self,
app_id,
user_id,
category_id,
creditLine,
principal,
result,
threshold,
model_name,
is_preAudit,
table='jdx_withdraw_model_results'):
sql = "insert into {} " \
"(app_id, user_id, category_id, creditLine, principal, result, threshold, model, is_preAudit) values " \
"('{}', '{}', '{}', {}, {}, {}, {}, '{}',{})".format(table, app_id, user_id, category_id, creditLine,
principal, result, threshold, model_name,
is_preAudit)
return self.mysql_ml.update(sql) == 1
def save_op_socre_result(self,
app_id,
user_id,
target,
server,
category_id,
model,
model_score,
op_score,
table='jdx_op_score_results'):
sql = "insert into {} " \
"(app_id, user_id, target, server, category_id, model, model_score, op_score) " \
"values ('{}', '{}', '{}', '{}', '{}', '{}', {}, {})".format(table, app_id, user_id, target,
server, category_id, model, model_score,
op_score)
return self.mysql_ml.update(sql) == 1
def __init__(self):
super().__init__()
self.queue_flow = mm.get_queue(cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivate = Mongo(**cm.config['mongo_jd_cl'])
self.category_list = self.get_category_configs()
class JdxCategoryService(message_server.MessageServer):
def __init__(self):
super().__init__()
self.queue_flow = mm.get_queue(cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivate = Mongo(**cm.config['mongo_jd_cl'])
self.category_list = self.get_category_configs()
def handle_msg(self, msg_dict):
app_id = msg_dict['app_id']
user_id = msg_dict['user_id']
derivate_data = self.get_derivate_data(app_id)
channel = derivate_data.get('X_Origin_source')
workflow, flow_type = self.get_workflow(derivate_data)
category_id = self.get_category_id(derivate_data, workflow)
msg_out = {
"app_id": app_id,
"job_name": msg_dict['job_name'],
"category_id": category_id,
}
self.queue_flow.send_message(msg_out)
self.save_category_results(app_id, user_id, category_id, channel,
flow_type)
def get_category_configs(self, table='jdx_category_relations'):
sql = """SELECT category_id,relation from {} WHERE is_setup=1;""".format(
table)
category_list = self.mysql_ml.query(sql)
return category_list
def get_derivate_data(self, app_id, collection='derivables'):
der_dict = self.mongo_derivate.get_collection(collection).find_one(
{'_id': app_id.upper()})
return der_dict
def save_category_results(self,
app_id,
user_id,
category_id,
channel,
flow_type,
table='jdx_category_results'):
sql = """INSERT INTO {} (app_id,user_id,category_id,channel,flow_type) VALUES('{}','{}','{}','{}','{}');""".format(
table, app_id, user_id, category_id, channel, flow_type)
self.mysql_ml.update(sql)
def get_category_id(self, derivate_data, workflow):
aeval = Interpreter()
category_id = None
default_category_id = None
for category in self.category_list:
if category['relation'] == workflow + '_default':
default_category_id = category['category_id']
try:
relation_dict = json.loads(category['relation'])
flag = []
for k, v in relation_dict.items():
aeval.symtable['VALUE'] = derivate_data.get(k)
flag.append(aeval(v))
if all(flag):
category_id = category['category_id']
break
except:
pass
if not category_id:
category_id = default_category_id
return category_id
def get_workflow(self, derivate_data):
x_workflow = derivate_data['X_WorkFlow_flag']
mx_data = derivate_data.get('X_MX_RawReport')
if x_workflow == 'card':
workflow = 'opencard'
flow_type = 'c'
elif x_workflow == 'reloan' and mx_data:
workflow = 'reloanwithdraw_mx'
flow_type = 'w'
elif x_workflow == 'reloan' and not mx_data:
workflow = 'reloanwithdraw'
flow_type = 'w'
elif x_workflow == 'quota':
workflow = 'quota'
flow_type = 'q'
elif x_workflow == 'firstloan':
workflow = 'firstwithdraw'
flow_type = 'f'
return workflow, flow_type
def __init__(self):
super().__init__()
self.queue_flow = mm.get_queue(cm.config['queue']['flow'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
class JDXMLCreditServer(message_server.MessageServer):
def __init__(self):
super(JDXMLCreditServer, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_der = Mongo(**cm.config['mongo_jd_cl'])
self.model_config_list = self.get_model_list()
self.target = 'jd'
def handle_msg(self, msg_dict):
if 'app_id' not in msg_dict:
logging.warning('No key in msg called appId!')
app_id = str(msg_dict['app_id']).lower()
user_id = msg_dict['user_id']
category_id = msg_dict['category_id']
model = self.choose_model(category_id, self.target)
model_score = self.run_model(app_id, model)
op_score = self.get_op_score(model_score, model['op_coef'],
model['op_intercept'])
self.save_model_result(app_id, user_id, -1, -1, {500: model_score}, '',
1, -1, model['model_name'], category_id)
self.save_op_socre_result(app_id, user_id, self.target, 'open_card',
category_id, model['model_name'],
model_score, op_score)
msg = {
"app_id": app_id,
"job_name": msg_dict['job_name'],
"model_name": model['model_name'],
"op_score": op_score,
"model_score": model_score
}
self.queue_flow.send_message(msg)
def choose_model(self, category_id, target):
models = []
for model in self.model_config_list:
if model['category_id'] == category_id and model[
'target'] == target:
models.append(model)
rand = np.random.random()
model_prod = 0
for model in models:
model_prod += model['prob']
if rand < model_prod:
return model
return None
def get_model_list(self, table='jdx_category_model_relations'):
sql = """select category_id,model_name,target,estor_path,op_coef,op_intercept,prob from {} where is_setup=1;""".format(
table)
model_list = self.mysql_ml.query(sql)
return model_list
def run_model(self, app_id, chosed_model):
"""跑模型，获得模型分数"""
model_path = chosed_model['estor_path']
feature_data = self.get_der_data(app_id)
with open(model_path, 'rb') as f:
pk_obj = pickle.load(f)
model = pk_obj['model']
agg_attr = []
fields = model.get_params()['enum'].clean_col_names
for attr in fields:
agg_attr.append(feature_data.get(attr))
# use pandas to format data, the dimension should be 1 x len(fields)
result_df = pd.DataFrame(agg_attr, index=fields).T
assert result_df.ndim == 2, 'result_df should be two dimension'
# transform data and make prediction
pred_probas = model.predict_proba(result_df)
model_score = pred_probas[:, 1][0]
return float(model_score)
def get_der_data(self, app_id, collection='derivables'):
"""获取特征数据"""
result = {}
try:
result = self.mongo_der.get_collection(collection).find({
'_id':
app_id.upper()
}).next()
result = {key: value for key, value in result.items()}
except Exception:
pass
return result
def get_op_score(self, model_score, op_coef, op_intercept):
if not (op_coef and op_intercept):
return 0
model_ln_odds = math.log(model_score / (1 - model_score)) / math.log(2)
ln_real_odds = op_coef * model_ln_odds + op_intercept
op_score = max(min(60 * ln_real_odds + 300, 1000), 0)
return op_score
def save_model_result(self,
app_id,
user_id,
credit_ml,
credit_final,
result,
threshold,
judge_by,
credit_by,
model_name,
category_id,
table='machine_learning_jdcl_results'):
sql = "insert into {} " \
"(app_id, user_id, credit_ml, credit_final, result, threshold, judge_by, credit_by, model, category_id) " \
"values ('{}', '{}', {}, {}, '{}', '{}', {}, {}, '{}', '{}')".format(table, app_id, user_id,
credit_ml,
credit_final, json.dumps(result),
threshold, judge_by,
credit_by, model_name, category_id)
return self.mysql_ml.update(sql) == 1
def save_op_socre_result(self,
app_id,
user_id,
target,
server,
category_id,
model,
model_score,
op_score,
table='jdx_op_score_results'):
sql = "insert into {} " \
"(app_id, user_id, target, server, category_id, model, model_score, op_score) " \
"values ('{}', '{}', '{}', '{}', '{}', '{}', {}, {})".format(table, app_id, user_id, target,
server, category_id, model, model_score,
op_score)
return self.mysql_ml.update(sql) == 1
class JDXModifyQuota(message_server.MessageServer):
def __init__(self):
super(JDXModifyQuota, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivable = Mongo(**cm.config['mongo_jd_cl'])
self.quota_modify_strategy_list = self.get_quota_modify_strategy_list(
) # 获取全部额度变更策略配置
def handle_msg(self, msg_dict):
app_id = msg_dict['app_id']
category_id = msg_dict['category_id']
model_score = msg_dict['model_score'] # 预测的模型分
model_name = msg_dict['model_name'] # 预测的模型名
data = self.get_user_data_by_app_id(app_id)
user_id = data.get('X_UserId')
try:
original_quota = int(float(data.get('X_JD_CreditLine'))) # 用户当前额度
original_principal = int(float(
data.get('X_pre_app_principal'))) # 用户最后一次的提现金额
quota_modify_count = int(
data.get('X_credit_adjustment_times')) # 提额次数
# 根据模型分和当前额度选择一个额度变更策略
quota_modify_strategy = self.get_quota_modify_strategy(
category_id, model_score, original_quota)
# 根据额度变更策略确定额度的变化标签
modify_tag = self.get_modify_tag(quota_modify_strategy)
# 最终额度
if modify_tag >= 0:
final_quota = round(original_quota +
original_principal * modify_tag)
else:
final_quota = round(original_quota * (1 + modify_tag))
# 用户年龄小于22时，额度打折
try:
if int(data['X_IdCardAge']) < 22:
final_quota = round(final_quota * 0.5)
except:
pass
# 额度变化值
quota_variance = final_quota - original_quota
logging.info("app_id:{}, final_credit:{}".format(
app_id, final_quota))
self.queue_flow.send_message({
"app_id": app_id,
"job_name": msg_dict['job_name'],
"final_quota": str(final_quota)
})
# 保存额度变更结果
self.save_quota_modify_results(
app_id, user_id, category_id,
quota_modify_strategy['modify_name'], model_score, model_name,
original_quota, original_principal,
quota_modify_strategy['score_segment'],
quota_modify_strategy['quota_segment'],
quota_modify_strategy['level'], quota_variance, final_quota,
quota_modify_count)
except:
logging.warning(traceback.format_exc())
final_quota = int(float(data.get('X_JD_CreditLine')))
quota_variance = 0
logging.info("app_id:{}, final_credit:{}".format(
app_id, final_quota))
self.queue_flow.send_message({
"app_id": app_id,
"job_name": msg_dict['job_name'],
"final_quota": str(final_quota)
})
# 保存额度变更结果
self.save_quota_modify_results(
app_id, user_id, category_id, '', model_score, model_name,
data.get('X_JD_CreditLine'), -1, '', '', 0, quota_variance,
final_quota, data.get('X_credit_adjustment_times'))
def get_quota_modify_strategy(self, category_id, model_score,
original_quota):
for quota_modify_strategy in self.quota_modify_strategy_list:
score_segment = json.loads(quota_modify_strategy['score_segment'])
quota_segment = json.loads(quota_modify_strategy['quota_segment'])
score_in = model_score in Interval(score_segment[0],
score_segment[1],
upper_closed=False)
quota_in = original_quota in Interval(quota_segment[0],
quota_segment[1],
upper_closed=False)
if quota_modify_strategy[
'category_id'] == category_id and score_in and quota_in:
return quota_modify_strategy
def get_modify_tag(self, quota_modify_strategy):
action_space = json.loads(quota_modify_strategy['action_space'])
action_prob = json.loads(quota_modify_strategy['action_prob'])
rand = np.random.random()
modify_prob = 0
for i, prob in enumerate(action_prob):
modify_prob += prob
if rand < modify_prob:
modify_tag = action_space[i]
return modify_tag
def save_quota_modify_results(self,
app_id,
user_id,
category_id,
modify_name,
model_score,
model_name,
original_quota,
original_principal,
score_segment,
quota_segment,
level,
quota_variance,
final_quota,
quota_modify_count,
table='jdx_quota_modify_results'):
sql = "insert into {} (app_id,user_id,category_id,modify_name,model_score,model_name,original_quota," \
"original_principal,score_segment,quota_segment,level,quota_variance,final_quota,quota_modify_count) " \
"values ('{}', '{}', '{}', '{}', {}, '{}',{},{}, '{}','{}',{},{},{},{})". \
format(table, app_id, user_id, category_id, modify_name, model_score, model_name, original_quota,
original_principal,score_segment, quota_segment,level,quota_variance, final_quota, quota_modify_count)
self.mysql_ml.update(sql)
def get_quota_modify_strategy_list(self, table='jdx_quota_modify_configs'):
sql = "select modify_name,category_id,score_segment,quota_segment,level,action_space,action_prob " \
"from {} where is_setup=1;" \
.format(table)
quota_modify_strategy_list = self.mysql_ml.query(sql)
return quota_modify_strategy_list
def get_user_data_by_app_id(self, app_id, collection='derivables'):
key_lists = [
'X_UserId', 'X_JD_CreditLine', 'X_pre_app_principal',
'X_credit_adjustment_times', 'X_IdCardAge'
]
fields = {x: 1 for x in key_lists}
data = self.mongo_derivable.get_collection(collection).find_one(
{'_id': app_id.upper()}, fields)
return data
class JDXMLQuotaServer(message_server.MessageServer):
def __init__(self):
super(JDXMLQuotaServer, self).__init__()
self.queue_flow = message_manager.get_queue(
cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivable = Mongo(**cm.config['mongo_jd_cl'])
self.model_config_list = self.get_model_list()
self.target = 'jd'
def handle_msg(self, msg_dict):
if 'app_id' not in msg_dict:
logging.warning('No key in msg called appId!')
app_id = str(msg_dict['app_id']).lower()
category_id = msg_dict['category_id']
model = self.choose_model(category_id, self.target)
model_score = self.run_model(app_id, model)
msg = {
"app_id": app_id,
"job_name": msg_dict['job_name'],
"model_name": model['model_name'],
"model_score": model_score
}
self.queue_flow.send_message(msg)
def choose_model(self, category_id, target):
models = []
for model in self.model_config_list:
if model['category_id'] == category_id and model[
'target'] == target:
models.append(model)
rand = np.random.random()
model_prod = 0
for model in models:
model_prod += model['prob']
if rand < model_prod:
return model
return None
def get_model_list(self, table='jdx_category_model_relations'):
sql = """select category_id,model_name,target,estor_path,op_coef,op_intercept,prob from {} where is_setup=1;""".format(
table)
model_list = self.mysql_ml.query(sql)
return model_list
def run_model(self, app_id, chosed_model):
"""跑模型，获得模型分数"""
model_path = chosed_model['estor_path']
feature_data = self.get_der_data(app_id)
with open(model_path, 'rb') as f:
pk_obj = pickle.load(f)
model = pk_obj['model']
agg_attr = []
fields = model.get_params()['enum'].clean_col_names
for attr in fields:
agg_attr.append(feature_data.get(attr))
# use pandas to format data, the dimension should be 1 x len(fields)
result_df = pd.DataFrame(agg_attr, index=fields).T
assert result_df.ndim == 2, 'result_df should be two dimension'
# transform data and make prediction
pred_probas = model.predict_proba(result_df)
model_score = pred_probas[:, 1][0]
return float(model_score)
def get_der_data(self, app_id, collection='derivables'):
"""获取特征数据"""
result = {}
try:
result = self.mongo_derivable.get_collection(collection).find({
'_id':
app_id.upper()
}).next()
result = {key: value for key, value in result.items()}
except Exception:
pass
return result
class JdxRuler(MessageServer):
def __init__(self):
super().__init__()
self.queue_flow = mm.get_queue(cm.config['queue_flow']['name'])
self.mysql_ml = MySql(**cm.config['mysql_jd_cl'])
self.mongo_derivate = Mongo(**cm.config['mongo_jd_cl'])
self.rule_config_list = self.get_rule_list()
self.group_tag_config_list = self.get_group_tag_list()
def handle_msg(self, msg_dict):
app_id = str(msg_dict['app_id'])
user_id = msg_dict.get('user_id', '')
category_id = msg_dict.get('category_id', '')
der_data = self.get_der_data(app_id)
group_tag, flow_type = self.get_group_tag_and_flow_type(der_data) # 控制组标签
rule_check_result = self.get_rule_check_result(category_id, flow_type, group_tag, der_data)
hit_rule_str = "!".join(rule_check_result['hit_rule_code'])
msg = {
"app_id": app_id,
"job_name": msg_dict['job_name'],
"group_tag": group_tag,
"rule_pass": rule_check_result['result'],
"hit_rule_str": hit_rule_str
}
self.queue_flow.send_message(msg)
saved_data = {
'app_id': app_id,
'user_id': user_id,
'category_id': category_id,
'hit_rule_code': rule_check_result['hit_rule_code'],
'actual_hit_rule_code': rule_check_result['actual_hit_rule_code'],
'result': rule_check_result['result'],
'actual_result': rule_check_result['actual_result'],
'group_tag': group_tag,
'flow_type': flow_type
}
self.save_rule_result(saved_data)
def get_rule_list(self, table='jd_ml_rules'):
"""获取所有规则配置信息"""
sql = """select category_id,rule_code,pass_prob,rule_params,flow_type,group_tag from {} where is_setup=1;""".format(
table)
rule_list = self.mysql_ml.query(sql)
return rule_list
def get_der_data(self, app_id, collection='derivables'):
"""获取特征数据"""
result = {}
try:
result = self.mongo_derivate.get_collection(collection).find({'_id': app_id.upper()}).next()
result = {key: value for key, value in result.items()}
except Exception:
pass
return result
def save_rule_result(self, saved_data, table='jd_ml_rule_check_result'):
"""保存规则检查结果"""
app_id = saved_data.get('app_id')
user_id = saved_data.get('user_id', '')
category_id = saved_data.get('category_id')
hit_rule_code = json.dumps(saved_data.get('hit_rule_code'))
actual_hit_rule_code = json.dumps(saved_data.get('actual_hit_rule_code'))
result = saved_data.get('result')
actual_result = saved_data.get('actual_result')
group_tag = saved_data.get('group_tag')
flow_type = saved_data.get('flow_type')
sql = "insert into {} (app_id,user_id,category_id,hit_rule_code,actual_hit_rule_code,result,actual_result," \
"group_tag,rule_type) values ('{}','{}','{}','{}','{}',{},{},'{}','{}')".format(table, app_id, user_id,
category_id,
hit_rule_code,
actual_hit_rule_code,
result, actual_result,
group_tag, flow_type)
self.mysql_ml.update(sql)
def get_rule_check_result(self, category_id, flow_type, group_tag, der_data):
"""得到规则检测结果"""
if category_id not in [rule['category_id'] for rule in self.rule_config_list]:
if flow_type == 'c':
category_id = 'default_opencard'
elif flow_type == 'f':
category_id = 'default_firstloan'
elif flow_type == 'w':
category_id = 'default_reloan'
rule_list = []
for rule_config in self.rule_config_list:
if rule_config['category_id'] == category_id and rule_config['flow_type'] == flow_type and rule_config[
'group_tag'] == group_tag:
rule_list.append(rule_config)
hit_rule_code = []
actual_hit_rule_code = []
for rule in rule_list:
rule_code = rule['rule_code']
pass_prob = int(rule['pass_prob'])
rule_params = rule['rule_params']
rule_params = json.loads(rule_params)
check_list = []
for rule_dict in rule_params['con']:
try:
f_value = der_data.get(rule_dict['field'])
s_value = rule_dict.get('val')
op = rule_dict.get('op')
flag = self.__condition_exec(f_value, s_value, op)
except:
flag = False
check_list.append(flag)
check_result = False
if rule_params['log_op'] == 'and':
check_result = all(check_list)
if rule_params['log_op'] == 'or':
check_result = any(check_list)
if check_result:
hit_rule_code.append(rule_code)
actual_hit_rule_code.append(rule_code)
prob = np.random.random()
if prob < pass_prob:
hit_rule_code.remove(rule_code)
result = 0 if hit_rule_code else 1
actual_result = 0 if actual_hit_rule_code else 1
rule_check_result = {
'hit_rule_code': hit_rule_code,
'actual_hit_rule_code': actual_hit_rule_code,
'result': result,
'actual_result': actual_result,
}
return rule_check_result
def get_group_tag_list(self, table='jdx_proba_tags'):
"""获取控制组"""
sql = """select tag,proba,flow_type from {} where is_setup=1;""".format(table)
group_tag_list = self.mysql_ml.query(sql)
return group_tag_list
def get_group_tag_and_flow_type(self,der_data):
X_WorkFlow_flag = der_data.get('X_WorkFlow_flag')
if X_WorkFlow_flag == 'card':
flow_type = 'c'
elif X_WorkFlow_flag == 'firstloan':
flow_type = 'f'
elif X_WorkFlow_flag == 'reloan':
flow_type = 'w'
for tag_dict in self.group_tag_config_list:
if tag_dict['flow_type'] == flow_type:
tag_config = tag_dict
prob = np.random.random()
# B表示为控制组
if prob < float(tag_config['proba']):
group_tag = 'B'
else:
group_tag = 'A'
return group_tag, flow_type
def __condition_exec(self, f_value, s_value, op):
result = False
if op == '==':
result = f_value == s_value
elif op == '>':
result = float(f_value) > float(s_value)
elif op == '>=':
result = float(f_value) >= float(s_value)
elif op == '<':
result = float(f_value) < float(s_value)
elif op == '<=':
result = float(f_value) <= float(s_value)
elif op == '!=':
result = f_value != s_value
elif op == 'in':
result = f_value in s_value
elif op == 'nin':
result = f_value not in s_value
return result