def test_eval_with_limits_holdout_timeout_with_results_in_queue(
self, pynisher_mock):
def side_effect(**kwargs):
queue = kwargs['queue']
queue.put({
'status': StatusType.SUCCESS,
'loss': 0.5,
'additional_run_info': {}
})
m1 = unittest.mock.Mock()
m2 = unittest.mock.Mock()
m1.return_value = m2
pynisher_mock.return_value = m1
m2.side_effect = side_effect
m2.exit_status = pynisher.TimeoutException
m2.wall_clock_time = 30
# Test for a succesful run
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
info = ta.start(None, instance=None, cutoff=30)
self.assertEqual(info[0], StatusType.SUCCESS)
self.assertEqual(info[1], 0.5)
self.assertIsInstance(info[2], float)
# And a crashed run which is in the queue
def side_effect(**kwargs):
queue = kwargs['queue']
queue.put({
'status': StatusType.CRASHED,
'loss': 2.0,
'additional_run_info': {}
})
m2.side_effect = side_effect
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
info = ta.start(None, instance=None, cutoff=30)
self.assertEqual(info[0], StatusType.CRASHED)
self.assertEqual(info[1], 1.0)
self.assertIsInstance(info[2], float)
def test_eval_with_limits_holdout_timeout_with_results_in_queue(self, pynisher_mock):
config = unittest.mock.Mock()
config.config_id = 198
def side_effect(**kwargs):
queue = kwargs['queue']
queue.put({'status': StatusType.SUCCESS,
'loss': 0.5,
'additional_run_info': {}})
m1 = unittest.mock.Mock()
m2 = unittest.mock.Mock()
m1.return_value = m2
pynisher_mock.return_value = m1
m2.side_effect = side_effect
m2.exit_status = pynisher.TimeoutException
m2.wall_clock_time = 30
# Test for a succesful run
ta = ExecuteTaFuncWithQueue(backend=self.backend, autosklearn_seed=1,
port=self.logger_port,
resampling_strategy='holdout',
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
pynisher_context='forkserver',
)
info = ta.run_wrapper(RunInfo(config=config, cutoff=30, instance=None,
instance_specific=None, seed=1, capped=False))
self.assertEqual(info[1].status, StatusType.SUCCESS)
self.assertEqual(info[1].cost, 0.5)
self.assertIsInstance(info[1].time, float)
self.assertNotIn('exitcode', info[1].additional_info)
# And a crashed run which is in the queue
def side_effect(**kwargs):
queue = kwargs['queue']
queue.put({'status': StatusType.CRASHED,
'loss': 2.0,
'additional_run_info': {}})
m2.side_effect = side_effect
ta = ExecuteTaFuncWithQueue(backend=self.backend, autosklearn_seed=1,
port=self.logger_port,
resampling_strategy='holdout',
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
pynisher_context='forkserver',
)
info = ta.run_wrapper(RunInfo(config=config, cutoff=30, instance=None,
instance_specific=None, seed=1, capped=False))
self.assertEqual(info[1].status, StatusType.CRASHED)
self.assertEqual(info[1].cost, 1.0)
self.assertIsInstance(info[1].time, float)
self.assertNotIn('exitcode', info[1].additional_info)
def test_cutoff_lower_than_remaining_time(self, pynisher_mock):
config = unittest.mock.Mock()
config.config_id = 198
ta = ExecuteTaFuncWithQueue(
backend=self.backend,
autosklearn_seed=1,
port=self.logger_port,
resampling_strategy='holdout',
stats=self.stats,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
pynisher_context='forkserver',
)
self.stats.ta_runs = 1
ta.run_wrapper(
RunInfo(config=config,
cutoff=30,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(pynisher_mock.call_args[1]['wall_time_in_s'], 4)
self.assertIsInstance(pynisher_mock.call_args[1]['wall_time_in_s'],
int)
def test_eval_with_limits_holdout_fail_timeout(self, pynisher_mock):
config = unittest.mock.Mock()
config.config_id = 198
m1 = unittest.mock.Mock()
m2 = unittest.mock.Mock()
m1.return_value = m2
pynisher_mock.return_value = m1
m2.exit_status = pynisher.TimeoutException
m2.wall_clock_time = 30
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=30,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[1].status, StatusType.TIMEOUT)
self.assertEqual(info[1].cost, 1.0)
self.assertIsInstance(info[1].time, float)
def test_eval_with_limits_holdout_2(self, eval_houldout_mock):
def side_effect(*args, **kwargs):
queue = kwargs['queue']
queue.put({
'status': StatusType.SUCCESS,
'loss': 0.5,
'additional_run_info': kwargs['instance']
})
eval_houldout_mock.side_effect = side_effect
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
self.scenario.wallclock_limit = 180
instance = "{'subsample': 30}"
info = ta.start(None, cutoff=30, instance=instance)
self.assertEqual(info[0], StatusType.SUCCESS)
self.assertEqual(info[-1], {
'message': "{'subsample': 30}",
'configuration_origin': 'UNKNOWN'
})
def test_eval_with_limits_holdout_fail_silent(self, pynisher_mock):
pynisher_mock.return_value = None
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
# The following should not fail because abort on first config crashed is false
info = ta.start(config=None, instance=None, cutoff=60)
self.assertEqual(info[0], StatusType.CRASHED)
self.assertEqual(info[1], 1.0)
self.assertIsInstance(info[2], float)
self.assertEqual(info[3], {
'configuration_origin': 'UNKNOWN',
'error': "Result queue is empty"
})
self.stats.ta_runs += 1
info = ta.start(config=None, instance=None, cutoff=30)
self.assertEqual(info[0], StatusType.CRASHED)
self.assertEqual(info[1], 1.0)
self.assertIsInstance(info[2], float)
self.assertEqual(info[3], {
'configuration_origin': 'UNKNOWN',
'error': "Result queue is empty"
})
def test_exception_in_target_function(self, eval_holdout_mock):
config = unittest.mock.Mock()
config.config_id = 198
eval_holdout_mock.side_effect = ValueError
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
self.stats.submitted_ta_runs += 1
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=30,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[1].status, StatusType.CRASHED)
self.assertEqual(info[1].cost, 1.0)
self.assertIsInstance(info[1].time, float)
self.assertEqual(info[1].additional_info['error'], 'ValueError()')
self.assertIn('traceback', info[1].additional_info)
def test_eval_with_limits_holdout(self, pynisher_mock):
pynisher_mock.side_effect = safe_eval_success_mock
config = unittest.mock.Mock()
config.config_id = 198
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=30,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[0].config.config_id, 198)
self.assertEqual(info[1].status, StatusType.SUCCESS)
self.assertEqual(info[1].cost, 0.5)
self.assertIsInstance(info[1].time, float)
def test_eval_with_limits_holdout_2(self, eval_houldout_mock):
config = unittest.mock.Mock()
config.config_id = 198
def side_effect(*args, **kwargs):
queue = kwargs['queue']
queue.put({'status': StatusType.SUCCESS,
'loss': 0.5,
'additional_run_info': kwargs['instance']})
eval_houldout_mock.side_effect = side_effect
ta = ExecuteTaFuncWithQueue(backend=self.backend, autosklearn_seed=1,
port=self.logger_port,
resampling_strategy='holdout',
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
pynisher_context='fork',
)
self.scenario.wallclock_limit = 180
instance = "{'subsample': 30}"
info = ta.run_wrapper(RunInfo(config=config, cutoff=30, instance=instance,
instance_specific=None, seed=1, capped=False))
self.assertEqual(info[1].status, StatusType.SUCCESS)
self.assertEqual(len(info[1].additional_info), 2)
self.assertIn('configuration_origin', info[1].additional_info)
self.assertEqual(info[1].additional_info['message'], "{'subsample': 30}")
def test_eval_with_limits_holdout_fail_memory_error(self, pynisher_mock):
pynisher_mock.side_effect = MemoryError
config = unittest.mock.Mock()
config.config_id = 198
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=log_loss,
cost_for_crash=get_cost_of_crash(log_loss),
abort_on_first_run_crash=False,
)
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=30,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[1].status, StatusType.MEMOUT)
# For logloss, worst possible result is MAXINT
worst_possible_result = MAXINT
self.assertEqual(info[1].cost, worst_possible_result)
self.assertIsInstance(info[1].time, float)
def test_eval_with_limits_holdout_fail_silent(self, pynisher_mock):
pynisher_mock.return_value = None
config = unittest.mock.Mock()
config.origin = 'MOCK'
config.config_id = 198
ta = ExecuteTaFuncWithQueue(
backend=self.backend,
autosklearn_seed=1,
port=self.logger_port,
resampling_strategy='holdout',
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
pynisher_context='fork',
)
# The following should not fail because abort on first config crashed is false
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=60,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[1].status, StatusType.CRASHED)
self.assertEqual(info[1].cost, 1.0)
self.assertIsInstance(info[1].time, float)
self.assertEqual(
info[1].additional_info, {
'configuration_origin': 'MOCK',
'error': "Result queue is empty",
'exit_status': 0,
'exitcode': 0,
'subprocess_stdout': '',
'subprocess_stderr': ''
})
self.stats.submitted_ta_runs += 1
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=30,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[1].status, StatusType.CRASHED)
self.assertEqual(info[1].cost, 1.0)
self.assertIsInstance(info[1].time, float)
self.assertEqual(
info[1].additional_info, {
'configuration_origin': 'MOCK',
'error': "Result queue is empty",
'exit_status': 0,
'exitcode': 0,
'subprocess_stdout': '',
'subprocess_stderr': ''
})
def test_cutoff_lower_than_remaining_time(self, pynisher_mock):
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
self.stats.ta_runs = 1
ta.start(None, cutoff=30, instance=None)
self.assertEqual(pynisher_mock.call_args[1]['wall_time_in_s'], 4)
self.assertIsInstance(pynisher_mock.call_args[1]['wall_time_in_s'],
int)
def test_zero_or_negative_cutoff(self, pynisher_mock):
config = unittest.mock.Mock()
config.config_id = 198
ta = ExecuteTaFuncWithQueue(backend=self.backend, autosklearn_seed=1,
port=self.logger_port,
resampling_strategy='holdout',
stats=self.stats,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
self.scenario.wallclock_limit = 5
self.stats.submitted_ta_runs += 1
run_info, run_value = ta.run_wrapper(RunInfo(config=config, cutoff=9, instance=None,
instance_specific=None, seed=1, capped=False))
self.assertEqual(run_value.status, StatusType.STOP)
def test_eval_with_limits_holdout(self, pynisher_mock):
pynisher_mock.side_effect = safe_eval_success_mock
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
info = ta.start(None, instance=None, cutoff=30)
self.assertEqual(info[0], StatusType.SUCCESS)
self.assertEqual(info[1], 0.5)
self.assertIsInstance(info[2], float)
def test_zero_or_negative_cutoff(self, pynisher_mock):
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
self.scenario.wallclock_limit = 5
self.stats.ta_runs += 1
self.assertRaises(BudgetExhaustedException,
ta.start,
None,
instance=None,
cutoff=9)
def test_silent_exception_in_target_function(self):
config = unittest.mock.Mock()
config.config_id = 198
delattr(self.backend, 'save_targets_ensemble')
ta = ExecuteTaFuncWithQueue(
backend=self.backend,
port=self.logger_port,
autosklearn_seed=1,
resampling_strategy='holdout',
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
iterative=False,
pynisher_context='fork',
)
ta.pynisher_logger = unittest.mock.Mock()
self.stats.submitted_ta_runs += 1
info = ta.run_wrapper(
RunInfo(config=config,
cutoff=3000,
instance=None,
instance_specific=None,
seed=1,
capped=False))
self.assertEqual(info[1].status,
StatusType.CRASHED,
msg=str(info[1].additional_info))
self.assertEqual(info[1].cost, 1.0)
self.assertIsInstance(info[1].time, float)
self.assertIn(
info[1].additional_info['error'],
("""AttributeError("'BackendMock' object has no attribute """
"""'save_targets_ensemble'",)""",
"""AttributeError("'BackendMock' object has no attribute """
"""'save_targets_ensemble'")""",
"""AttributeError('save_targets_ensemble')"""))
self.assertNotIn('exitcode', info[1].additional_info)
self.assertNotIn('exit_status', info[1].additional_info)
self.assertNotIn('traceback', info[1])
def test_eval_with_limits_holdout_fail_memory_error(self, pynisher_mock):
pynisher_mock.side_effect = MemoryError
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=log_loss,
cost_for_crash=get_cost_of_crash(log_loss),
abort_on_first_run_crash=False,
)
info = ta.start(None, instance=None, cutoff=30)
self.assertEqual(info[0], StatusType.MEMOUT)
# For logloss, worst possible result is MAXINT
worst_possible_result = MAXINT
self.assertEqual(info[1], worst_possible_result)
self.assertIsInstance(info[2], float)
def test_exception_in_target_function(self, eval_holdout_mock):
eval_holdout_mock.side_effect = ValueError
ta = ExecuteTaFuncWithQueue(
backend=BackendMock(),
autosklearn_seed=1,
resampling_strategy='holdout',
logger=self.logger,
stats=self.stats,
memory_limit=3072,
metric=accuracy,
cost_for_crash=get_cost_of_crash(accuracy),
abort_on_first_run_crash=False,
)
self.stats.ta_runs += 1
info = ta.start(None, instance=None, cutoff=30)
self.assertEqual(info[0], StatusType.CRASHED)
self.assertEqual(info[1], 1.0)
self.assertIsInstance(info[2], float)
self.assertEqual(info[3]['error'], 'ValueError()')
self.assertIn('traceback', info[3])
# To avoid the output "first run crashed"...
stats.submitted_ta_runs += 1
stats.finished_ta_runs += 1
memory_lim = memory_limit_factor * automl_arguments['memory_limit']
ta = ExecuteTaFuncWithQueue(
backend=automl.automl_._backend,
autosklearn_seed=seed,
resampling_strategy='test',
memory_limit=memory_lim,
disable_file_output=True,
logger=logger,
stats=stats,
scoring_functions=scoring_functions,
include=include,
metric=automl_arguments['metric'],
cost_for_crash=get_cost_of_crash(automl_arguments['metric']),
abort_on_first_run_crash=False,
)
run_info, run_value = ta.run_wrapper(
RunInfo(
config=config,
instance=None,
instance_specific=None,
seed=1,
cutoff=per_run_time_limit * 3,
capped=False,
))
if run_value.status == StatusType.SUCCESS:
assert len(
run_value.additional_info) > 1, run_value.additional_info
def __init__(
self,
config_space,
dataset_name,
backend,
total_walltime_limit,
func_eval_time_limit,
memory_limit,
metric,
watcher,
n_jobs,
dask_client: dask.distributed.Client,
start_num_run=1,
data_memory_limit=None,
num_metalearning_cfgs=25,
config_file=None,
seed=1,
metadata_directory=None,
resampling_strategy='holdout',
resampling_strategy_args=None,
include_estimators=None,
exclude_estimators=None,
include_preprocessors=None,
exclude_preprocessors=None,
disable_file_output=False,
smac_scenario_args=None,
get_smac_object_callback=None,
scoring_functions=None,
ensemble_callback: typing.Optional[EnsembleBuilderManager] = None,
):
super(AutoMLSMBO, self).__init__()
# data related
self.dataset_name = dataset_name
self.datamanager = None
self.metric = metric
self.task = None
self.backend = backend
# the configuration space
self.config_space = config_space
# the number of parallel workers/jobs
self.n_jobs = n_jobs
self.dask_client = dask_client
# Evaluation
self.resampling_strategy = resampling_strategy
if resampling_strategy_args is None:
resampling_strategy_args = {}
self.resampling_strategy_args = resampling_strategy_args
# and a bunch of useful limits
self.worst_possible_result = get_cost_of_crash(self.metric)
self.total_walltime_limit = int(total_walltime_limit)
self.func_eval_time_limit = int(func_eval_time_limit)
self.memory_limit = memory_limit
self.data_memory_limit = data_memory_limit
self.watcher = watcher
self.num_metalearning_cfgs = num_metalearning_cfgs
self.config_file = config_file
self.seed = seed
self.metadata_directory = metadata_directory
self.start_num_run = start_num_run
self.include_estimators = include_estimators
self.exclude_estimators = exclude_estimators
self.include_preprocessors = include_preprocessors
self.exclude_preprocessors = exclude_preprocessors
self.disable_file_output = disable_file_output
self.smac_scenario_args = smac_scenario_args
self.get_smac_object_callback = get_smac_object_callback
self.scoring_functions = scoring_functions
self.ensemble_callback = ensemble_callback
dataset_name_ = "" if dataset_name is None else dataset_name
logger_name = '%s(%d):%s' % (self.__class__.__name__, self.seed,
":" + dataset_name_)
self.logger = get_logger(logger_name)
