def __init__(self, max_iter_num, dataset_id, ns_num, **kwargs):
use_log = (kwargs.get("use_log", True), )
shuffle = kwargs.get("shuffle", False)
vw_format = kwargs.get("vw_format", True)
print("dataset_id", dataset_id)
self.vw_examples, self.Y = get_data(
max_iter_num,
dataset_id=dataset_id,
vw_format=vw_format,
max_ns_num=ns_num,
shuffle=shuffle,
use_log=use_log,
)
self.max_iter_num = min(max_iter_num, len(self.Y))
self._problem_info = {
"max_iter_num": self.max_iter_num,
"dataset_id": dataset_id,
"ns_num": ns_num,
}
self._problem_info.update(kwargs)
self._fixed_hp_config = kwargs.get("fixed_hp_config", {})
self.namespace_feature_dim = AutoVW.get_ns_feature_dim_from_vw_example(
self.vw_examples[0])
self._raw_namespaces = list(self.namespace_feature_dim.keys())
self._setup_search()
def __init__(self, max_iter_num, dataset_id, ns_num, **kwargs):
super().__init__(max_iter_num, dataset_id, ns_num, **kwargs)
from flaml.onlineml import VowpalWabbitTrial
self._raw_namespaces = list(AutoVW.get_ns_feature_dim_from_vw_example(self.vw_examples[0]).keys())
self._info_key_list = ["dataset_id", "max_iter_num", "ns_num", "shuffle", "use_log"]
self.problem_id = 'vw-ns-interaction-' + ('_').join(
[str(self._problem_info.get(k, 'None')) for k in self._info_key_list])
self._setup_search()
logger.info('search space %s %s', self._search_space, self.problem_id)
def test_supervised_vw_tune_namespace(self):
# basic experiment setting
vw_oml_problem_args, vw_online_aml_problem = get_vw_tuning_problem()
autovw = AutoVW(max_live_model_num=5,
search_space=vw_online_aml_problem.search_space,
init_config=vw_online_aml_problem.init_config,
min_resource_lease='auto',
random_seed=2345)
cumulative_loss_list = online_learning_loop(
vw_online_aml_problem.max_iter_num,
vw_online_aml_problem.vw_examples,
autovw,
loss_func=vw_oml_problem_args["fixed_hp_config"].get(
"loss_function", "squared"),
)
print('final average loss:',
sum(cumulative_loss_list) / len(cumulative_loss_list))
def __init__(self, max_iter_num, dataset_id, ns_num, **kwargs):
use_log = kwargs.get('use_log', True),
shuffle = kwargs.get('shuffle', False)
vw_format = kwargs.get('vw_format', True)
print('dataset_id', dataset_id)
self.vw_examples, self.Y = get_data(max_iter_num,
dataset_id=dataset_id,
vw_format=vw_format,
max_ns_num=ns_num,
shuffle=shuffle,
use_log=use_log)
self.max_iter_num = min(max_iter_num, len(self.Y))
self._problem_info = {
'max_iter_num': self.max_iter_num,
'dataset_id': dataset_id,
'ns_num': ns_num,
}
self._problem_info.update(kwargs)
self._fixed_hp_config = kwargs.get('fixed_hp_config', {})
self.namespace_feature_dim = AutoVW.get_ns_feature_dim_from_vw_example(
self.vw_examples[0])
self._raw_namespaces = list(self.namespace_feature_dim.keys())
self._setup_search()
}
# setup configs for the experiments to run
alg_dic = {}
for alg_name in args.method_list:
if alg_name in method_data.keys():
alg_args = method_data[alg_name]
alg_alias = '_'.join([alg_name, exp_config])
if 'is_naive' not in alg_args or not alg_args['is_naive']:
autovw_args = auto_alg_common_args.copy()
autovw_args.update(alg_args['config'])
# use the method_name+current_exp_config as the alias for the algorithm
logger.info('alg_alias %s %s', alg_alias, alg_args)
logger.info('trial runner config %s %s',
alg_args['config'], autovw_args)
alg_dic[alg_alias] = AutoVW(**autovw_args)
else:
vw_args = fixed_hp_config.copy()
if 'config' in alg_args and alg_args[
'config'] is not None:
vw_args.update(alg_args['config'])
alg_dic[alg_alias] = pyvw.vw(**vw_args)
else:
print('alg name not in methods.yaml')
NotImplementedError
# convert method names from input to the names in alg_dic
for alg_name, alg in alg_dic.items():
max_iter_num = vw_online_aml_problem.max_iter_num
time_start = time.time()
print('----------running', alg_name, '-----------')
### get result file name