def updata_jobtable(tocken,un,pa):
#同步云脑数据库job信息
job = models.User_Job.objects.all().order_by("id")
job = job.exclude(state="STOPPED").exclude(state="FAIL").exclude(state="SUCCEEDED")
for jd in job:
print(jd)
jd_detail = API_tools.get_jobinfo(jd.jobid,tocken,un,pa)
if jd_detail["code"] == "S000":
jd.state = jd_detail["payload"]["jobStatus"]["state"]
timeStamp2 = int(jd_detail['payload']['jobStatus']["completedTime"])
if timeStamp2 != 0:
timeArray2 = time.localtime(timeStamp2 / 1000)
otherStyleTime2 = time.strftime("%Y-%m-%d %H:%M:%S", timeArray2)
jd.completedTime = otherStyleTime2
jd.save()
print("$$$$$$$$ Update Dataset Success")
def insert_train_job(user,job_id,algo=None):
res_info=API_tools.get_jobinfo(job_id,"",user.username,user.first_name)
job=res_info["payload"]
print("RESPONSE JOB INFO: ",res_info)
jid=job['id']
if jid != job_id:
return ["Job id Error"]
createdTime = datetime.datetime.fromtimestamp(job['jobStatus']["completedTime"]/1000)
# compeletedTime = createdTime
models.customize_job.objects.create(
job_id=jid,
name=job['name'],
state=job['jobStatus']['state'],
created_at=createdTime,
# completed_at=compeletedTime,
uid=user,
algo_id=algo,
)
return []
def refresh_train_job(user):
'''
Args:
user: 用户实例
'''
# 以本地数据库的信息为主
un=user.username
pa=user.first_name
token=API_tools.get_tocken(un,pa)
job_list=models.customize_job.objects.all()
for rec in job_list:
# print("JOB: ", rec.jobid)
info=API_tools.get_jobinfo(rec.job_id,token,un,pa)
if info['code'] == 'S000': # 存在该任务
job=info['payload']
rec.state = job["jobStatus"]["state"]
timeStamp2 = int(job['jobStatus']["completedTime"])
if timeStamp2 == 0: continue
rec.completed_at = datetime.datetime.fromtimestamp(
timeStamp2/1000 )
rec.save()
else:
rec=models.customize_job.objects.get(id=rec.id)
rec.delete()
def AutoSearch(method:str,hyper:dict,package:dict,api_config:dict,interval:int=60):
print("子进程（%s） 开始执行，父进程为（%s）" % (os.getpid(), os.getppid()))
time.sleep(5)
print("========MISSION REPORT========")
print("pid: %s METHOD is: "%(os.getpid()),method)
time.sleep(5)
print("-"*24)
time.sleep(5)
print("pid: %s Hypers: "%(os.getpid()))
a=[print(k,v) for k,v in enumerate(hyper.items())]
print("-"*24)
time.sleep(5)
print("pid: %s Job config is: "%(os.getpid()))
a=[print(k,v) for k,v in package.items()]
print("-"*24)
time.sleep(5)
print("pid: %s Search config is: "%(os.getpid()))
a=[print(k,v) for k,v in api_config.items()]
print("="*24)
return
optimizer=None # 【需要字典形式列出的黑盒方法集合】
for ii in range(hyper['epoch']):
points=optimizer.suggest(hyper['suggest'])
success=[]
jobid=[]
for jj,point in enumerate(points):
# 创建每个建议参数的任务
print(point)
param=get_param(package['param'],point) # 【需要知道point的形式】
info=API_tools.mission_submit(
job_name="as_"+package['job_name'],
project_dir=package['project_dir'],
main_file = package['main_file'],
param=param,
resource=package['resource'],
username=package['username'],
password=package['password'],)
if type(info) != dict or info.__contains__("code") and info["code"]!="S000":
print("ERROR: mission creation failed.\n* Epoch:{}, P:{}, {}".format(ii,jj,point))
success.append(False)
else:
jobid.append(info['payload']['jobId'])
success.append(True)
if not all(success):
print("Some mission failed, exit this search procedure.")
break
finish=[False]*hyper['suggest']
retry=[10]*hyper['suggest']
stop=False
while(all(finish) or stop):
for i,jid in enumerate(jobid):
info=API_tools.get_jobinfo(jid,username=package['username'],password=package['password'])
if info['code'] == 'S000':
status=info['payload']['jobStatus']['state'].lower()
if status == 'succeeded':
finish[i] = True
if status == 'waiting':
pass
if status == 'running':
pass
if status == 'failed':
stop=True
if status == 'stopped':
stop=True
else:
retry[i]-=1
if stop : break
stop=any([x<0 for x in retry])
time.sleep(interval)
if all(finish) and not stop:
ylist=get_result(package['username'],hyper['result'])
optimizer.observe(points,ylist)