def _fit(self, datamanager):
# Reset learnt stuff
self.models_ = None
self.ensemble_ = None
# Check arguments prior to doing anything!
if self._resampling_strategy not in ['holdout', 'holdout-iterative-fit',
'cv', 'nested-cv', 'partial-cv']:
raise ValueError('Illegal resampling strategy: %s' %
self._resampling_strategy)
if self._resampling_strategy == 'partial-cv' and \
self._ensemble_size != 0:
raise ValueError("Resampling strategy partial-cv cannot be used "
"together with ensembles.")
self._backend._make_internals_directory()
if self._keep_models:
try:
os.mkdir(self._backend.get_model_dir())
except OSError:
self._logger.warning("model directory already exists")
if not self._shared_mode:
raise
self._metric = datamanager.info['metric']
self._task = datamanager.info['task']
self._label_num = datamanager.info['label_num']
set_auto_seed(self._seed)
# == Pickle the data manager, here, because no more global
# OneHotEncoding
data_manager_path = self._backend.save_datamanager(datamanager)
self._save_ensemble_data(
datamanager.data['X_train'],
datamanager.data['Y_train'])
time_for_load_data = self._stopwatch.wall_elapsed(self._dataset_name)
if self._debug_mode:
self._print_load_time(
self._dataset_name,
self._time_for_task,
time_for_load_data,
self._logger)
# == Perform dummy predictions
if self._resampling_strategy in ['holdout', 'holdout-iterative-fit']:
self._do_dummy_prediction(datamanager)
# = Create a searchspace
# Do this before One Hot Encoding to make sure that it creates a
# search space for a dense classifier even if one hot encoding would
# make it sparse (tradeoff; if one hot encoding would make it sparse,
# densifier and truncatedSVD would probably lead to a MemoryError,
# like this we can't use some of the preprocessing methods in case
# the data became sparse)
self.configuration_space, configspace_path = _create_search_space(
self._tmp_dir,
datamanager.info,
self._backend,
self._stopwatch,
self._logger,
self._include_estimators,
self._include_preprocessors)
self.configuration_space_created_hook(datamanager)
# == RUN ensemble builder
# Do this before calculating the meta-features to make sure that the
# dummy predictions are actually included in the ensemble even if
# calculating the meta-features takes very long
proc_ensembles = self.run_ensemble_builder()
# == Calculate metafeatures
meta_features = _calculate_metafeatures(
data_feat_type=datamanager.feat_type,
data_info_task=datamanager.info['task'],
x_train=datamanager.data['X_train'],
y_train=datamanager.data['Y_train'],
basename=self._dataset_name,
watcher=self._stopwatch,
metalearning_cnt=self._initial_configurations_via_metalearning,
logger=self._logger)
self._stopwatch.start_task('OneHot')
datamanager.perform1HotEncoding()
self._stopwatch.stop_task('OneHot')
if meta_features is None:
initial_configurations = []
elif datamanager.info['task'] in [MULTICLASS_CLASSIFICATION,
BINARY_CLASSIFICATION,
MULTILABEL_CLASSIFICATION]:
meta_features_encoded = _calculate_metafeatures_encoded(
self._dataset_name,
datamanager.data['X_train'],
datamanager.data['Y_train'],
self._stopwatch,
self._logger)
self._logger.debug(meta_features.__repr__(verbosity=2))
self._logger.debug(meta_features_encoded.__repr__(verbosity=2))
initial_configurations = _get_initial_configuration(
meta_features,
meta_features_encoded,
self._dataset_name,
self._metric,
self.configuration_space,
self._task,
self._metadata_directory,
self._initial_configurations_via_metalearning,
datamanager.info[
'is_sparse'],
self._stopwatch,
self._logger)
_print_debug_info_of_init_configuration(
initial_configurations,
self._dataset_name,
self._time_for_task,
self._logger,
self._stopwatch)
else:
initial_configurations = []
self._logger.warning('Metafeatures encoded not calculated')
# == RUN SMAC
if (datamanager.info["task"] == BINARY_CLASSIFICATION) or \
(datamanager.info["task"] == MULTICLASS_CLASSIFICATION):
config = {'balancing:strategy': 'weighting',
'classifier:__choice__': 'sgd',
'classifier:sgd:loss': 'hinge',
'classifier:sgd:penalty': 'l2',
'classifier:sgd:alpha': 0.0001,
'classifier:sgd:fit_intercept': 'True',
'classifier:sgd:n_iter': 5,
'classifier:sgd:learning_rate': 'optimal',
'classifier:sgd:eta0': 0.01,
'classifier:sgd:average': 'True',
'imputation:strategy': 'mean',
'one_hot_encoding:use_minimum_fraction': 'True',
'one_hot_encoding:minimum_fraction': 0.1,
'preprocessor:__choice__': 'no_preprocessing',
'rescaling:__choice__': 'min/max'}
elif datamanager.info["task"] == MULTILABEL_CLASSIFICATION:
config = {'classifier:__choice__': 'adaboost',
'classifier:adaboost:algorithm': 'SAMME.R',
'classifier:adaboost:learning_rate': 1.0,
'classifier:adaboost:max_depth': 1,
'classifier:adaboost:n_estimators': 50,
'balancing:strategy': 'weighting',
'imputation:strategy': 'mean',
'one_hot_encoding:use_minimum_fraction': 'True',
'one_hot_encoding:minimum_fraction': 0.1,
'preprocessor:__choice__': 'no_preprocessing',
'rescaling:__choice__': 'none'}
else:
config = None
self._logger.info("Tasktype unknown: %s" %
TASK_TYPES_TO_STRING[datamanager.info["task"]])
if config is not None:
try:
configuration = Configuration(self.configuration_space, config)
config_string = convert_conf2smac_string(configuration)
initial_configurations = [config_string] + initial_configurations
except ValueError:
pass
# == RUN SMAC
proc_smac = run_smac(tmp_dir=self._tmp_dir, basename=self._dataset_name,
time_for_task=self._time_for_task,
ml_memory_limit=self._ml_memory_limit,
data_manager_path=data_manager_path,
configspace_path=configspace_path,
initial_configurations=initial_configurations,
per_run_time_limit=self._per_run_time_limit,
watcher=self._stopwatch, backend=self._backend,
seed=self._seed,
resampling_strategy=self._resampling_strategy,
resampling_strategy_arguments=self._resampling_strategy_arguments,
shared_mode=self._shared_mode)
procs = []
if proc_smac is not None:
procs.append(proc_smac)
if proc_ensembles is not None:
procs.append(proc_ensembles)
if self._queue is not None:
self._queue.put([time_for_load_data, data_manager_path, procs])
else:
for proc in procs:
proc.wait()
# Delete AutoSklearn environment variable
del_auto_seed()
# In case
try:
del self._datamanager
except Exception:
pass
if self._queue is None:
self._load_models()
return self
def _fit(self, datamanager):
# Reset learnt stuff
self.models_ = None
self.ensemble_indices_ = None
# Check arguments prior to doing anything!
if self._resampling_strategy not in ['holdout', 'holdout-iterative-fit',
'cv', 'nested-cv', 'partial-cv']:
raise ValueError('Illegal resampling strategy: %s' %
self._resampling_strategy)
if self._resampling_strategy == 'partial-cv' and \
self._ensemble_size != 0:
raise ValueError("Resampling strategy partial-cv cannot be used "
"together with ensembles.")
self._backend._make_internals_directory()
if self._keep_models:
try:
os.mkdir(self._backend.get_model_dir())
except OSError:
self._logger.warning("model directory already exists")
if not self._shared_mode:
raise
self._metric = datamanager.info['metric']
self._task = datamanager.info['task']
self._label_num = datamanager.info['label_num']
set_auto_seed(self._seed)
# == Pickle the data manager, here, because no more global
# OneHotEncoding
data_manager_path = self._backend.save_datamanager(datamanager)
self._save_ensemble_data(
datamanager.data['X_train'],
datamanager.data['Y_train'])
time_for_load_data = self._stopwatch.wall_elapsed(self._dataset_name)
if self._debug_mode:
self._print_load_time(
self._dataset_name,
self._time_for_task,
time_for_load_data,
self._logger)
# == Perform dummy predictions
self._do_dummy_prediction(datamanager)
# = Create a searchspace
# Do this before One Hot Encoding to make sure that it creates a
# search space for a dense classifier even if one hot encoding would
# make it sparse (tradeoff; if one hot encoding would make it sparse,
# densifier and truncatedSVD would probably lead to a MemoryError,
# like this we can't use some of the preprocessing methods in case
# the data became sparse)
self.configuration_space, configspace_path = _create_search_space(
self._tmp_dir,
datamanager.info,
self._backend,
self._stopwatch,
self._logger,
self._include_estimators,
self._include_preprocessors)
self.configuration_space_created_hook(datamanager)
# == Calculate metafeatures
meta_features = _calculate_metafeatures(
data_feat_type=datamanager.feat_type,
data_info_task=datamanager.info['task'],
x_train=datamanager.data['X_train'],
y_train=datamanager.data['Y_train'],
basename=self._dataset_name,
watcher=self._stopwatch,
metalearning_cnt=self._initial_configurations_via_metalearning,
logger=self._logger)
self._stopwatch.start_task('OneHot')
datamanager.perform1HotEncoding()
self._stopwatch.stop_task('OneHot')
if meta_features is None:
initial_configurations = []
elif datamanager.info['task'] in [MULTICLASS_CLASSIFICATION,
BINARY_CLASSIFICATION,
MULTILABEL_CLASSIFICATION]:
meta_features_encoded = _calculate_metafeatures_encoded(
self._dataset_name,
datamanager.data['X_train'],
datamanager.data['Y_train'],
self._stopwatch,
self._logger)
self._logger.debug(meta_features.__repr__(verbosity=2))
self._logger.debug(meta_features_encoded.__repr__(verbosity=2))
initial_configurations = _get_initial_configuration(
meta_features,
meta_features_encoded,
self._dataset_name,
self._metric,
self.configuration_space,
self._task,
self._metadata_directory,
self._initial_configurations_via_metalearning,
datamanager.info[
'is_sparse'],
self._stopwatch,
self._logger)
_print_debug_info_of_init_configuration(
initial_configurations,
self._dataset_name,
self._time_for_task,
self._logger,
self._stopwatch)
else:
initial_configurations = []
self._logger.warning('Metafeatures encoded not calculated')
# == RUN SMAC
if (datamanager.info["task"] == BINARY_CLASSIFICATION) or \
(datamanager.info["task"] == MULTICLASS_CLASSIFICATION):
config = {'balancing:strategy': 'weighting',
'classifier:__choice__': 'sgd',
'classifier:sgd:loss': 'hinge',
'classifier:sgd:penalty': 'l2',
'classifier:sgd:alpha': 0.0001,
'classifier:sgd:fit_intercept': 'True',
'classifier:sgd:n_iter': 5,
'classifier:sgd:learning_rate': 'optimal',
'classifier:sgd:eta0': 0.01,
'classifier:sgd:average': 'True',
'imputation:strategy': 'mean',
'one_hot_encoding:use_minimum_fraction': 'True',
'one_hot_encoding:minimum_fraction': 0.1,
'preprocessor:__choice__': 'no_preprocessing',
'rescaling:__choice__': 'min/max'}
elif datamanager.info["task"] == MULTILABEL_CLASSIFICATION:
config = {'classifier:__choice__': 'adaboost',
'classifier:adaboost:algorithm': 'SAMME.R',
'classifier:adaboost:learning_rate': 1.0,
'classifier:adaboost:max_depth': 1,
'classifier:adaboost:n_estimators': 50,
'balancing:strategy': 'weighting',
'imputation:strategy': 'mean',
'one_hot_encoding:use_minimum_fraction': 'True',
'one_hot_encoding:minimum_fraction': 0.1,
'preprocessor:__choice__': 'no_preprocessing',
'rescaling:__choice__': 'none'}
else:
config = None
self._logger.info("Tasktype unknown: %s" %
TASK_TYPES_TO_STRING[datamanager.info["task"]])
if config is not None:
configuration = Configuration(self.configuration_space, config)
config_string = convert_conf2smac_string(configuration)
initial_configurations = [config_string] + initial_configurations
# == RUN SMAC
proc_smac = run_smac(tmp_dir=self._tmp_dir, basename=self._dataset_name,
time_for_task=self._time_for_task,
ml_memory_limit=self._ml_memory_limit,
data_manager_path=data_manager_path,
configspace_path=configspace_path,
initial_configurations=initial_configurations,
per_run_time_limit=self._per_run_time_limit,
watcher=self._stopwatch, backend=self._backend,
seed=self._seed,
resampling_strategy=self._resampling_strategy,
resampling_strategy_arguments=self._resampling_strategy_arguments,
shared_mode=self._shared_mode)
# == RUN ensemble builder
proc_ensembles = self.run_ensemble_builder()
procs = []
if proc_smac is not None:
procs.append(proc_smac)
if proc_ensembles is not None:
procs.append(proc_ensembles)
if self._queue is not None:
self._queue.put([time_for_load_data, data_manager_path, procs])
else:
for proc in procs:
proc.wait()
# Delete AutoSklearn environment variable
del_auto_seed()
# In case
try:
del self._datamanager
except Exception:
pass
if self._queue is None:
self._load_models()
return self
def _fit(self, datamanager):
# Reset learnt stuff
self.models_ = None
self.ensemble_indices_ = None
# Check arguments prior to doing anything!
if self._resampling_strategy not in ['holdout', 'holdout-iterative-fit',
'cv', 'nested-cv', 'partial-cv']:
raise ValueError('Illegal resampling strategy: %s' %
self._resampling_strategy)
if self._resampling_strategy == 'partial-cv' and \
self._ensemble_size != 0:
raise ValueError("Resampling strategy partial-cv cannot be used "
"together with ensembles.")
self._backend._make_internals_directory()
if self._keep_models:
try:
os.mkdir(self._backend.get_model_dir())
except OSError:
self._logger.warn("model directory already exists")
if not self._shared_mode:
raise
self._metric = datamanager.info['metric']
self._task = datamanager.info['task']
self._label_num = datamanager.info['label_num']
set_auto_seed(self._seed)
# == Pickle the data manager, here, because no more global
# OneHotEncoding
data_manager_path = self._backend.save_datamanager(datamanager)
self._save_ensemble_data(
datamanager.data['X_train'],
datamanager.data['Y_train'])
time_for_load_data = self._stopwatch.wall_elapsed(self._dataset_name)
if self._debug_mode:
self._print_load_time(
self._dataset_name,
self._time_for_task,
time_for_load_data,
self._logger)
# == Perform dummy predictions
self._do_dummy_prediction(datamanager)
# = Create a searchspace
# Do this before One Hot Encoding to make sure that it creates a
# search space for a dense classifier even if one hot encoding would
# make it sparse (tradeoff; if one hot encoding would make it sparse,
# densifier and truncatedSVD would probably lead to a MemoryError,
# like this we can't use some of the preprocessing methods in case
# the data became sparse)
self.configuration_space, configspace_path = _create_search_space(
self._tmp_dir,
datamanager.info,
self._backend,
self._stopwatch,
self._logger,
self._include_estimators,
self._include_preprocessors)
self.configuration_space_created_hook(datamanager)
# == Calculate metafeatures
meta_features = _calculate_metafeatures(
data_feat_type=datamanager.feat_type,
data_info_task=datamanager.info['task'],
x_train=datamanager.data['X_train'],
y_train=datamanager.data['Y_train'],
basename=self._dataset_name,
watcher=self._stopwatch,
metalearning_cnt=self._initial_configurations_via_metalearning,
logger=self._logger)
self._stopwatch.start_task('OneHot')
datamanager.perform1HotEncoding()
self._stopwatch.stop_task('OneHot')
if meta_features is None:
initial_configurations = []
elif datamanager.info['task'] in [MULTICLASS_CLASSIFICATION,
BINARY_CLASSIFICATION]:
meta_features_encoded = _calculate_metafeatures_encoded(
self._dataset_name,
datamanager.data['X_train'],
datamanager.data['Y_train'],
self._stopwatch,
self._logger)
self._logger.debug(meta_features.__repr__(verbosity=2))
self._logger.debug(meta_features_encoded.__repr__(verbosity=2))
initial_configurations = _get_initial_configuration(
meta_features,
meta_features_encoded,
self._dataset_name,
self._metric,
self.configuration_space,
self._task,
self._metadata_directory,
self._initial_configurations_via_metalearning,
datamanager.info[
'is_sparse'],
self._stopwatch,
self._logger)
_print_debug_info_of_init_configuration(
initial_configurations,
self._dataset_name,
self._time_for_task,
self._logger,
self._stopwatch)
else:
initial_configurations = []
self._logger.warn('Metafeatures encoded not calculated')
# == RUN SMACtmp_dir, basename, time_for_task, ml_memory_limit,
proc_smac = run_smac(tmp_dir=self._tmp_dir, basename=self._dataset_name,
time_for_task=self._time_for_task,
ml_memory_limit=self._ml_memory_limit,
data_manager_path=data_manager_path,
configspace_path=configspace_path,
initial_configurations=initial_configurations,
per_run_time_limit=self._per_run_time_limit,
watcher=self._stopwatch, backend=self._backend,
seed=self._seed,
resampling_strategy=self._resampling_strategy,
resampling_strategy_arguments=self._resampling_strategy_arguments,
shared_mode=self._shared_mode)
# == RUN ensemble builder
proc_ensembles = self.run_ensemble_builder()
procs = []
if proc_smac is not None:
procs.append(proc_smac)
if proc_ensembles is not None:
procs.append(proc_ensembles)
if self._queue is not None:
self._queue.put([time_for_load_data, data_manager_path, procs])
else:
for proc in procs:
proc.wait()
# Delete AutoSklearn environment variable
del_auto_seed()
# In case
try:
del self._datamanager
except Exception:
pass
if self._queue is None:
self._load_models()
return self