def run(self, X, y):
super(Joint, self).run(X, y)
current_pipeline_configuration = {}
current_algo_configuration = {}
trials = Trials()
algorithm = self.config['algorithm']
space = {
'pipeline': self.PIPELINE_SPACE,
'algorithm': ALGORITHM_SPACE.get_domain_space(algorithm),
}
obj_pl = functools.partial(objective_joint,
algorithm=self.config['algorithm'],
X=X,
y=y,
context=self.context,
config=self.config)
fmin(
fn=obj_pl,
space=space,
algo=tpe.suggest,
max_evals=None,
max_time=self.config['time'],
trials=trials,
show_progressbar=False,
verbose=0
)
best_config = self.context['best_config']
super(Joint, self).display_step_results(best_config)
current_pipeline_configuration = best_config['pipeline']
def run(self, X, y):
super(Iterative, self).run(X, y)
ranges = [
i for i in range(
0, self.config['time'] /
(self.config['step_pipeline'] + self.config['step_algorithm']))
]
current_pipeline_configuration = {}
current_algo_configuration = {}
trials_pipelines = Trials()
trials_algo = Trials()
reset_trial = False
for r in ranges:
print('## Data Pipeline')
if reset_trial:
trials_pipelines = Trials()
obj_pl = functools.partial(
objective_pipeline,
current_algo_config=current_algo_configuration,
algorithm=self.config['algorithm'],
X=X,
y=y,
context=self.context,
config=self.config)
fmin(fn=obj_pl,
space=self.PIPELINE_SPACE,
algo=tpe.suggest,
max_evals=None,
max_time=self.config['step_pipeline'],
trials=trials_pipelines,
show_progressbar=False,
verbose=0)
best_config = self.context['best_config']
current_pipeline_configuration = best_config['pipeline']
super(Iterative, self).display_step_results(best_config)
print('## Algorithm')
if reset_trial:
trials_algo = Trials()
obj_algo = functools.partial(
objective_algo,
current_pipeline_config=current_pipeline_configuration,
algorithm=self.config['algorithm'],
X=X,
y=y,
context=self.context,
config=self.config)
fmin(fn=obj_algo,
space=ALGORITHM_SPACE.get_domain_space(
self.config['algorithm']),
algo=tpe.suggest,
max_evals=None,
max_time=self.config['step_algorithm'],
trials=trials_algo,
show_progressbar=False,
verbose=0)
best_config = self.context['best_config']
current_pipeline_configuration = best_config['pipeline']
super(Iterative, self).display_step_results(best_config)
def run(self, X, y):
super(Adaptive, self).run(X, y)
current_pipeline_configuration = {}
current_algo_configuration = {}
trials_pipelines = Trials()
trials_algo = Trials()
timeout = False
start = time.time()
remaining = self.config['time']
while not timeout:
ellapsed = time.time() - start
remaining = self.config['time'] - ellapsed
timeout = remaining < 0
print('## Data Pipeline')
step_start = time.time()
if self.__reset_trials('pipeline'):
trials_pipelines = Trials()
obj_pl = functools.partial(
objective_pipeline,
current_algo_config=current_algo_configuration,
algorithm=self.config['algorithm'],
X=X,
y=y,
context=self.context,
config=self.config)
try:
fmin(fn=obj_pl,
space=PIPELINE_SPACE,
algo=tpe.suggest,
max_evals=None,
max_time=self.current_steptime['pipeline'],
trials=trials_pipelines,
show_progressbar=False,
verbose=0)
except Exception as e:
pass
best_config = self.context['best_config']
current_pipeline_configuration = best_config['pipeline']
super(Adaptive, self).display_step_results(best_config)
self.__determine_last_step(step_start)
self.__score_last_step()
self.__update_timestep('pipeline')
print('## Algorithm')
step_start = time.time()
if self.__reset_trials('algorithm'):
trials_algo = Trials()
obj_algo = functools.partial(
objective_algo,
current_pipeline_config=current_pipeline_configuration,
algorithm=self.config['algorithm'],
X=X,
y=y,
context=self.context,
config=self.config)
try:
fmin(fn=obj_algo,
space=ALGORITHM_SPACE.get_domain_space(
self.config['algorithm']),
algo=tpe.suggest,
max_evals=None,
max_time=self.current_steptime['algorithm'],
trials=trials_algo,
show_progressbar=False,
verbose=0)
except Exception as e:
pass
best_config = self.context['best_config']
current_pipeline_configuration = best_config['pipeline']
super(Adaptive, self).display_step_results(best_config)
self.__determine_last_step(step_start)
self.__score_last_step()
self.__update_timestep('algorithm')
self.context['current_steptime'] = self.current_steptime
self.context['history_steps_score'] = self.history_steps_score