def test_hyperopt_parameters(use_hyperband: bool) -> None:
sampler = TPESampler(**TPESampler.hyperopt_parameters())
study = optuna.create_study(
sampler=sampler,
pruner=optuna.pruners.HyperbandPruner() if use_hyperband else None)
study.optimize(lambda t: t.suggest_uniform("x", 10, 20), n_trials=50)
def test_hyperparam_optim_tpe():
# Define trainenv
train_env = GridWorld()
# Parameters
params = {"n_episodes": 500}
# Run AgentStats
stats_agent = AgentStats(DummyAgent, train_env, init_kwargs=params,
n_fit=4, eval_horizon=10, n_jobs=1)
# test hyperparameter optimization with TPE sampler
# using hyperopt default values
sampler_kwargs = TPESampler.hyperopt_parameters()
stats_agent.optimize_hyperparams(sampler_kwargs=sampler_kwargs)
def __init__(self, frameworks,
model_types,
pre_proc=None,
proc_dict=None,
post_proc=None,
scoring=None,
cv=3,
agg_func=np.mean,
tol=1e-7,
max_iter=50,
max_fails=0,
time_limit=None,
study_name=None,
save_cv_preds=False,
pruner=LinearExtrapolationPruner(n_steps_back=2, n_steps_forward=15, percentage_from_best=90),
sampler=TPESampler(**TPESampler.hyperopt_parameters()),
storage=None,
refit=True,
n_jobs=1,
verbose=1,
random_state=None):
self.frameworks = frameworks if isinstance(frameworks, list) else [frameworks]
self.model_types = model_types if isinstance(model_types, list) else [model_types]
self.pre_proc = pre_proc,
self.proc_dict = proc_dict,
self.post_proc = post_proc,
self.scoring = scoring
self.cv = cv
self.agg_func = agg_func
self.refit_ = refit
self.tol = tol
self.max_iter = max_iter
self.time_limit = time_limit
self.max_fails = max_fails
self.study_name = study_name
self.save_cv_preds = save_cv_preds
self.pruner = pruner
self.sampler = sampler
# TODO: add CMA Sampler
self.storage = storage
self.n_jobs = n_jobs
self.verbose = verbose
self.random_state = random_state
self.best_pipeline_ = None
def __init__(self, argument, grid_search_space=None):
self.name = ''
self.argument = argument
self.grid_search_space = grid_search_space
if self.argument.sampler == "grid":
assert self.grid_search_space is not None, "grid search spaceを指定してください"
self.sampler = GridSampler(self.grid_search_space)
self.n_trials = 1
for value in self.grid_search_space.values():
self.n_trials *= len(value)
# トライアル回数制限
# if self.n_trials > self.argument.n_trials:
# self.n_trials = self.argument.n_trials
self.obj_func_name = self.objective_grid
elif self.argument.sampler == "random":
self.sampler = RandomSampler(seed=self.argument.seed)
self.n_trials = self.argument.n_trials
self.obj_func_name = self.objective_no_grid
else:
self.sampler = TPESampler(**TPESampler.hyperopt_parameters(),
seed=self.argument.seed)
self.n_trials = self.argument.n_trials
self.obj_func_name = self.objective_no_grid
if self.n_trials == 1:
try:
mlflow.set_experiment(self.argument.experiment)
except Exception as e:
print(e)
else:
try:
mlflow.set_experiment(
self.argument.experiment + "_" +
datetime.now().strftime('%Y%m%d_%H:%M:%S'))
except Exception as e:
print(e)
self.study = optuna.create_study(sampler=self.sampler)
def test_hyperparam_optim_tpe():
# Define trainenv
train_env = (GridWorld, {})
# Run AgentManager
stats_agent = AgentManager(
DummyAgent,
train_env,
fit_budget=1,
init_kwargs={},
eval_kwargs={"eval_horizon": 5},
n_fit=4,
)
# test hyperparameter optimization with TPE sampler
# using hyperopt default values
sampler_kwargs = TPESampler.hyperopt_parameters()
stats_agent.optimize_hyperparams(sampler_kwargs=sampler_kwargs, n_trials=5)
stats_agent.clear_output_dir()
def __init__(self,
estimators,
pipeline=None,
pipe_params=None,
scoring=None,
cv=3,
agg_func=np.mean,
refit=True,
tol=1e-5,
max_iter=50,
time_limit=None,
max_fails=3,
study_name=None,
save_cv_preds=False,
pruner=SuccessiveHalvingPruner(min_resource=3,
reduction_factor=3),
sampler=TPESampler(**TPESampler.hyperopt_parameters()),
storage=None,
n_jobs=1,
verbose=1,
random_state=None):
self.estimators = estimators
self.pipeline = pipeline
self.pipe_params = pipe_params
self.scoring = scoring
self.cv = cv
self.agg_func = agg_func
self.refit_ = refit
self.tol = tol
self.max_iter = max_iter
self.time_limit = time_limit
self.max_fails = max_fails
self.study_name = study_name
self.save_cv_preds = save_cv_preds
self.pruner = pruner
self.sampler = sampler
self.storage = storage
self.n_jobs = n_jobs
self.verbose = verbose
self.random_state = random_state
self.best_estimator_ = None
def test_hyperopt_parameters():
# type: () -> None
sampler = TPESampler(**TPESampler.hyperopt_parameters())
study = optuna.create_study(sampler=sampler)
study.optimize(lambda t: t.suggest_uniform('x', 10, 20), n_trials=50)
#y_train = y_train.iloc[X_train.index].reset_index(drop=True)
#X_train.reset_index(inplace=True, drop=True)
#pipeline - pipeline_objective OR regular pipeline
#consider making pre-made steps with best practices (FS, scaling, etc...) then add option to concat to make one pipeline
#pipeline = Pipeline([("sel", SelectPercentile(chi2))])
#pipe_params = {"sel__percentile" : optuna.distributions.IntUniformDistribution(1,100)}
pipeline = None
pipe_params = None
from optuna.samplers import TPESampler
sampler = TPESampler(**TPESampler.hyperopt_parameters(), seed=SEED)
# sampler = optuna.integration.CmaEsSampler(n_startup_trials=40,
# independent_sampler=TPESampler(**TPESampler.hyperopt_parameters()),
# warn_independent_sampling=False, seed=SEED)
#TODO: make sure uninstalled estimators don't show up
xgb_linear = XGBClassifierHypster(
booster_list=['gblinear'],
lr_decay=0.1,
n_iter_per_round=2
#,param_dict={#'subsample' : 0.9,
#'eta' : optuna.distributions.LogUniformDistribution(0.8, 1.0)
# }
)
