def get(self, request, nnid):
"""
Manage AutoML define information include (id, desc and etc ) \n
Structure : AutoML - NetID - NetVer(Auto Generated by GA) - NetBatch (auto generated on every batch) \n
(1) Define AutoML Graph definition \n
(2) Select Type of Data \n
(3) Select Type of Anal algorithm \n
(4) Select range of hyper parameters \n
(5) Run - AutoML \n
(6) Check result of each generation with UI/UX (<- for this step) \n
(7) Select Best model you want use and activate it \n
---
# Class Name : RunManagerAutoStat
# Description:
Get Result of Train for each population
"""
try:
aml = AutoMlCommon(nnid)
stat_info = aml.get_stat_obj(nnid)
return_data = stat_info
return Response(json.dumps(return_data))
except Exception as e:
return_data = {"status": "404", "result": str(e)}
return Response(json.dumps(return_data))
def get(self, request, nnid):
"""
Manage AutoML define information include (id, desc and etc ) \n
Structure : AutoML - NetID - NetVer(Auto Generated by GA) - NetBatch (auto generated on every batch) \n
(1) Define AutoML Graph definition \n
(2) Select Type of Data \n
(3) Select Type of Anal algorithm \n
(4) Select range of hyper parameters \n
(5) Run - AutoML \n
(6) Check result of each generation with UI/UX (<- for this step) \n
(7) Select Best model you want use and activate it \n
---
# Class Name : RunManagerAutoStat
# Description:
Get Result of Train for each population
"""
try:
aml = AutoMlCommon(nnid)
stat_info = aml.get_stat_obj(nnid)
return_data = stat_info
return Response(json.dumps(return_data))
except Exception as e:
return_data = {"status": "404", "result": str(e)}
return Response(json.dumps(return_data))
def save_summary(self, info):
"""
save best survived results
:return:
"""
self.summary['best'] = info
AutoMlCommon().update_stat_obj(self.nn_id, self.summary)
def __init__(self, nn_id):
"""
initialize parms need to run auto mode
:return:
"""
self.auto_ml_info = AutoMlCommon(nn_id)
self.conf_info = self.auto_ml_info.conf_info
self.train_info = self.auto_ml_info.train_info
self.stat_info = self.auto_ml_info.stat_info
def put(self, request, nnid):
"""
- desc ; update cnn configuration data
"""
try:
return_data = AutoMlCommon().update_conf_obj(nnid, request.data)
return Response(json.dumps(return_data))
except Exception as e:
return_data = {"status": "404", "result": str(e)}
return Response(json.dumps(return_data))
def __init__(self, nn_id):
"""
initialize parms need to run auto mode
:return:
"""
self.nn_id = nn_id
self.auto_ml_info = AutoMlCommon(nn_id)
self.conf_info = self.auto_ml_info.conf_info
self.parm_info = self.auto_ml_info.parm_info
self.stat_info = self.auto_ml_info.stat_info
self.net_type = self.auto_ml_info.net_type
self.summary = {}
self.summary['bygen'] = []
self.summary['best'] = []
self.celery_flag = settings.CELERY_FLAG
def __init__(self, nn_id):
"""
initialize parms need to run auto mode
:return:
"""
self.nn_id = nn_id
self.auto_ml_info = AutoMlCommon(nn_id)
self.conf_info = self.auto_ml_info.conf_info
self.parm_info = self.auto_ml_info.parm_info
self.stat_info = self.auto_ml_info.stat_info
self.net_type = self.auto_ml_info.net_type
self.summary = {}
self.summary['bygen'] = []
self.summary['best'] = []
self.debug_mode = False
def run(self):
"""
run automl
:return:
"""
try:
# get genetic algorithm hyper parameters
generation = self.parm_info.get('generation')
population = self.parm_info.get('population')
survive = self.parm_info.get('survive')
# define gene list
networks = []
# iterate generation
for idx in range(generation):
# create new generations
if (idx == 0):
gen_nets, ver_data_sets = self.create_networks(
idx, population)
networks = networks + gen_nets
AutoMlCommon().reset_stat_obj(self.nn_id)
else:
gen_nets, ver_data_sets = self.create_networks(
idx, population - survive)
networks = networks + gen_nets
# train & evaluate networks
initial = True if idx == 0 else False
networks = self.train_networks(networks, initial=initial)
# set each train set flag to fin
self.set_train_finish(ver_data_sets)
# network acc set
accArr = 0
for net in networks:
return_data = NNCommonManager().get_nn_batch_info(
net['nn_id'], net['nn_wf_ver_id'])
re_acc = return_data[0]['acc_info']['acc']
networks[accArr]['acc'] = float(re_acc[len(re_acc) - 1])
accArr += 1
# print(return_data)
# update traing progress
self.update_summary(networks, survive)
# sort & discard
networks = self.discard_inferior(networks, survive)
return networks
except Exception as e:
logging.error("Error on running AutoML alogorithm : {0}".format(e))
def put(self, request, nnid):
"""
Your docs
---
# Class Name (must be separated by `---`)
# Description:
- name: name
description: Foobar long description goes here
"""
try:
return_data = AutoMlCommon().update_conf_obj(nnid, request.data)
return Response(json.dumps(return_data))
except Exception as e:
return_data = {"status": "404", "result": str(e)}
return Response(json.dumps(return_data))
def update_summary(self, networks, survive):
"""
update summary info of processing genetic algorithm
arranged by generation (sequece of list means number of generation
each node contains extra info like survive or not
:param networks: networks for one generations
:param survive: number of gene to survive
:return: dict type result info with extra flag
"""
# Regression sorting order add
if self.net_type.rfind("_reg") != -1:
networksNew = copy.deepcopy(
sorted(networks, key=lambda x: x.get('acc'), reverse=False))
else:
networksNew = copy.deepcopy(
sorted(networks, key=lambda x: x.get('acc'), reverse=True))
# result = list(map(lambda x : self.set_value(x, 'survive', True) , networks[0:survive]))
# self.summary['best'] = result
# result = result + list(map(lambda x : self.set_value(x, 'survive', False) , networks[survive:]))
# self.summary['bygen'].append(result)
surcnt = 1
result = []
best = []
for net in networksNew:
if survive < surcnt:
net['survive'] = False
else:
net['survive'] = True
best.append(net)
if survive == surcnt:
self.summary['best'] = best
result.append(net)
surcnt += 1
self.summary['bygen'].append(result)
AutoMlCommon().update_stat_obj(self.nn_id, self.summary)
# Set Best Active
nnCommonManager = NNCommonManager()
input_data = {}
input_data['active_flag'] = "Y"
input_data['nn_wf_ver_id'] = networksNew[0].get('nn_wf_ver_id')
nnCommonManager.update_nn_wf_info(networksNew[0].get('nn_id'),
input_data)
def update_summary(self, networks, survive):
"""
update summary info of processing genetic algorithm
arranged by generation (sequece of list means number of generation
each node contains extra info like survive or not
:param networks: networks for one generations
:param survive: number of gene to survive
:return: dict type result info with extra flag
"""
networks = sorted(networks, key=lambda x: x.get('acc'), reverse=True)
result = list(
map(lambda x: self.set_value(x, 'survive', True),
networks[0:survive]))
self.summary['best'] = result
result = result + list(
map(lambda x: self.set_value(x, 'survive', False),
networks[survive:]))
self.summary['bygen'].append(result)
AutoMlCommon().update_stat_obj(self.nn_id, self.summary)
def run(self):
"""
run automl
:return:
"""
try:
# get genetic algorithm hyper parameters
generation = self.parm_info.get('generation')
population = self.parm_info.get('population')
survive = self.parm_info.get('survive')
# define gene list
networks = []
# iterate generation
for idx in range(generation):
# create new generations
if (idx == 0):
gen_nets, ver_data_sets = self.create_networks(
idx, population)
networks = networks + gen_nets
AutoMlCommon().reset_stat_obj(self.nn_id)
else:
gen_nets, ver_data_sets = self.create_networks(
idx, population - survive)
networks = networks + gen_nets
# train & evaluate networks
networks = self.train_networks(networks)
# set each train set flag to fin
self.set_train_finish(ver_data_sets)
# update traing progress
self.update_summary(networks, survive)
# sort & discard
networks = self.discard_inferior(networks, survive)
return networks
except Exception as e:
logging.error("Error on running AutoML alogorithm : {0}".format(e))
def put(self, request, nnid):
"""
Manage hyperparameter for GA algorithm like eval, population, survive etc
Structure : AutoML - NetID - NetVer(Auto Generated by GA) - NetBatch (auto generated on every batch) \n
(1) Define AutoML Graph definition \n
(2) Select Type of Data \n
(3) Select Type of Anal algorithm \n
(4) Select range of hyper parameters (<- for this step)\n
(5) Run - AutoML \n
(6) Check result of each generation with UI/UX \n
(7) Select Best model you want use and activate it \n
---
# Class Name : RunManagerAutoConf
# Description:
Modifiy hyperparameters for genetic algorithm itself, if it is not set
Genetic Algorithm will run with default parmas
"""
try:
return_data = AutoMlCommon().update_conf_obj(nnid, request.data)
return Response(json.dumps(return_data))
except Exception as e:
return_data = {"status": "404", "result": str(e)}
return Response(json.dumps(return_data))
