def run_exp(config_file, plot_timechops=True, run_exp=True, n_jobs=1):
if plot_timechops:
visualize_timechop(config_file)
config, sql_engine, proj_folder = setup_experiment(config_file)
if run_exp:
if n_jobs> 1:
experiment = MultiCoreExperiment(
config=config,
db_engine=sql_engine,
n_processes=n_jobs,
n_db_processes=n_jobs,
project_path=proj_folder,
replace=False,
cleanup=True
)
else:
experiment = SingleThreadedExperiment(
config=config,
db_engine=sql_engine,
project_path=proj_folder,
cleanup=True
)
st = time.time()
experiment.run()
en = time.time()
print('Took {} seconds to run the experiement'.format(en-st))
def experiment(self):
self.root.setup() # Loading configuration (if exists)
db_url = self.root.db_url
config = self._load_config()
db_engine = create_engine(db_url)
common_kwargs = {
"db_engine": db_engine,
"project_path": self.args.project_path,
"config": config,
"replace": self.args.replace,
"materialize_subquery_fromobjs": self.args.materialize_fromobjs,
"features_ignore_cohort": self.args.features_ignore_cohort,
"matrix_storage_class": self.matrix_storage_map[self.args.matrix_format],
"profile": self.args.profile,
"save_predictions": self.args.save_predictions,
"skip_validation": not self.args.validate
}
if self.args.n_db_processes > 1 or self.args.n_processes > 1:
experiment = MultiCoreExperiment(
n_db_processes=self.args.n_db_processes,
n_processes=self.args.n_processes,
**common_kwargs,
)
else:
experiment = SingleThreadedExperiment(**common_kwargs)
return experiment
def experiment(self):
self.root.setup() # Loading configuration (if exists)
db_url = self.root.db_url
config = yaml.load(self.args.config)
db_engine = create_engine(db_url)
common_kwargs = {
"db_engine":
db_engine,
"project_path":
self.args.project_path,
"config":
config,
"replace":
self.args.replace,
"matrix_storage_class":
self.matrix_storage_map[self.args.matrix_format],
}
if self.args.n_db_processes > 1 or self.args.n_processes > 1:
experiment = MultiCoreExperiment(
n_db_processes=self.args.n_db_processes,
n_processes=self.args.n_processes,
**common_kwargs,
)
else:
experiment = SingleThreadedExperiment(**common_kwargs)
return experiment
def triage(ctx, config_file, triage_db, replace, debug):
config_file = os.path.join(os.sep, "triage", "experiment_config",
config_file)
click.echo(f"Using the config file {config_file}")
with open(config_file) as f:
experiment_config = yaml.load(f)
click.echo(
f"The output (matrices and models) of this experiment will be stored in triage/output"
)
click.echo(f"Using data stored in {triage_db}")
click.echo(
f"The experiment will utilize any preexisting matrix or model: {not replace}"
)
click.echo(f"Creating experiment object")
experiment = SingleThreadedExperiment(
config=experiment_config,
db_engine=sqlalchemy.create_engine(triage_db),
model_storage_class=FSModelStorageEngine,
project_path='/triage/output',
replace=replace)
ctx.obj = experiment
if debug:
logging.basicConfig(level=logging.DEBUG)
click.echo("Debug enabled (Expect A LOT of output at the screen!!!)")
click.echo("Experiment loaded")
def triage(ctx, config_file, triage_db, replace):
config_file = os.path.join(os.sep, "triage", "experiment_config",
config_file)
click.echo(f"Using the config file {config_file}")
with open(config_file) as f:
experiment_config = yaml.load(f)
click.echo(
f"The output (matrices and models) of this experiment will be stored in triage/output"
)
click.echo(
f"The experiment will utilize any preexisting matrix or model: {not replace}"
)
click.echo(f"Creating experiment object")
experiment = SingleThreadedExperiment(
config=experiment_config,
db_engine=sqlalchemy.create_engine(triage_db),
model_storage_class=FSModelStorageEngine,
project_path='/triage/output',
replace=replace)
ctx.obj = experiment
click.echo("Experiment loaded")
def test_noload_if_wrong_version(self):
experiment_config = sample_config()
experiment_config["config_version"] = "v0"
with TemporaryDirectory() as temp_dir:
with self.assertRaises(ValueError):
SingleThreadedExperiment(
config=experiment_config,
db_engine=None,
project_path=os.path.join(temp_dir, "inspections"),
)
def test_profiling(db_engine):
populate_source_data(db_engine)
with TemporaryDirectory() as temp_dir:
project_path = os.path.join(temp_dir, "inspections")
SingleThreadedExperiment(config=sample_config(),
db_engine=db_engine,
project_path=project_path,
profile=True).run()
assert len(os.listdir(os.path.join(project_path,
"profiling_stats"))) == 1
def test_experiment_tracker_in_parts(test_engine, project_path):
experiment = SingleThreadedExperiment(
config=sample_config(),
db_engine=test_engine,
project_path=project_path,
)
experiment.generate_matrices()
experiment.train_and_test_models()
with scoped_session(test_engine) as session:
experiment_run = session.query(ExperimentRun).get(experiment.run_id)
assert experiment_run.start_method == "generate_matrices"
def prepare_experiment(config):
with testing.postgresql.Postgresql() as postgresql:
db_engine = create_engine(postgresql.url())
populate_source_data(db_engine)
with TemporaryDirectory() as temp_dir:
experiment = SingleThreadedExperiment(config=config,
db_engine=db_engine,
project_path=os.path.join(
temp_dir, 'inspections'),
cleanup=False)
yield experiment
def test_noload_if_wrong_version(self):
experiment_config = sample_config()
experiment_config['config_version'] = 'v0'
with TemporaryDirectory() as temp_dir:
with self.assertRaises(ValueError):
SingleThreadedExperiment(
config=experiment_config,
db_engine=None,
model_storage_class=FSModelStorageEngine,
project_path=os.path.join(temp_dir, 'inspections'),
)
def test_noload_if_wrong_version(self):
experiment_config = sample_config()
experiment_config["config_version"] = "v0"
with TemporaryDirectory() as temp_dir, mock.patch(
"triage.util.conf.open",
side_effect=open_side_effect) as mock_file:
with self.assertRaises(ValueError):
SingleThreadedExperiment(
config=experiment_config,
db_engine=None,
project_path=os.path.join(temp_dir, "inspections"),
)
def test_load_if_right_version(self):
experiment_config = sample_config()
experiment_config["config_version"] = CONFIG_VERSION
with testing.postgresql.Postgresql() as postgresql:
db_engine = create_engine(postgresql.url())
with TemporaryDirectory() as temp_dir:
experiment = SingleThreadedExperiment(
config=experiment_config,
db_engine=db_engine,
project_path=os.path.join(temp_dir, "inspections"),
)
assert isinstance(experiment, SingleThreadedExperiment)
def test_experiment_tracker_in_parts(test_engine, project_path):
with mock.patch("triage.util.conf.open",
side_effect=open_side_effect) as mock_file:
experiment = SingleThreadedExperiment(
config=sample_config(),
db_engine=test_engine,
project_path=project_path,
)
experiment.generate_matrices()
experiment.train_and_test_models()
with scoped_session(test_engine) as session:
experiment_run = session.query(TriageRun).get(experiment.run_id)
assert experiment_run.start_method == "generate_matrices"
def finished_experiment(shared_db_engine, shared_project_storage):
"""A successfully-run experiment. Its database schemas and project storage can be queried.
Returns: (triage.experiments.SingleThreadedExperiment)
"""
populate_source_data(shared_db_engine)
base_config = sample_config()
experiment = SingleThreadedExperiment(
base_config,
db_engine=shared_db_engine,
project_path=shared_project_storage.project_path
)
experiment.run()
return experiment
def test_load_if_right_version(self):
experiment_config = sample_config()
experiment_config["config_version"] = CONFIG_VERSION
with testing.postgresql.Postgresql() as postgresql, TemporaryDirectory(
) as temp_dir, mock.patch("triage.util.conf.open",
side_effect=open_side_effect) as mock_file:
db_engine = create_engine(postgresql.url())
experiment = SingleThreadedExperiment(
config=experiment_config,
db_engine=db_engine,
project_path=os.path.join(temp_dir, "inspections"),
)
assert isinstance(experiment, SingleThreadedExperiment)
def experiment(self):
self.root.setup() # Loading configuration (if exists)
db_url = self.root.db_url
config = self._load_config()
db_engine = create_engine(db_url)
common_kwargs = {
"db_engine":
db_engine,
"project_path":
self.args.project_path,
"config":
config,
"replace":
self.args.replace,
"materialize_subquery_fromobjs":
self.args.materialize_fromobjs,
"features_ignore_cohort":
self.args.features_ignore_cohort,
"matrix_storage_class":
self.matrix_storage_map[self.args.matrix_format],
"profile":
self.args.profile,
"save_predictions":
self.args.save_predictions,
"skip_validation":
not self.args.validate
}
logger.info(f"Setting up the experiment")
logger.info(f"Configuration file: {self.args.config}")
logger.info(f"Results will be stored in DB: {self.root.db_url}")
logger.info(f"Artifacts will be saved in {self.args.project_path}")
try:
if self.args.n_db_processes > 1 or self.args.n_processes > 1:
experiment = MultiCoreExperiment(
n_db_processes=self.args.n_db_processes,
n_processes=self.args.n_processes,
**common_kwargs,
)
logger.info(
f"Experiment will run in multi core mode using {self.args.n_processes} processes and {self.args.n_db_processes} db processes"
)
else:
experiment = SingleThreadedExperiment(**common_kwargs)
logger.info("Experiment will run in serial fashion")
return experiment
except Exception:
logger.exception("Error occurred while creating the experiment!")
logger.info(
f"Experiment [config file: {self.args.config}] failed at creation"
)
def test_profiling(db_engine):
populate_source_data(db_engine)
with TemporaryDirectory() as temp_dir, mock.patch(
"triage.util.conf.open",
side_effect=open_side_effect) as mock_file:
project_path = os.path.join(temp_dir, "inspections")
SingleThreadedExperiment(
config=sample_config(),
db_engine=db_engine,
project_path=project_path,
profile=True,
).run()
assert len(os.listdir(os.path.join(project_path,
"profiling_stats"))) == 1
def test_experiment_tracker_exception(db_engine, project_path):
experiment = SingleThreadedExperiment(
config=sample_config(),
db_engine=db_engine,
project_path=project_path,
)
# no source data means this should blow up
with pytest.raises(Exception):
experiment.run()
with scoped_session(db_engine) as session:
experiment_run = session.query(ExperimentRun).get(experiment.run_id)
assert experiment_run.current_status == ExperimentRunStatus.failed
assert isinstance(experiment_run.last_updated_time, datetime.datetime)
assert experiment_run.stacktrace
def test_load_if_right_version(self):
experiment_config = sample_config()
experiment_config['config_version'] = CONFIG_VERSION
with testing.postgresql.Postgresql() as postgresql:
db_engine = create_engine(postgresql.url())
ensure_db(db_engine)
with TemporaryDirectory() as temp_dir:
experiment = SingleThreadedExperiment(
config=experiment_config,
db_engine=db_engine,
model_storage_class=FSModelStorageEngine,
project_path=os.path.join(temp_dir, 'inspections'),
)
assert isinstance(experiment, SingleThreadedExperiment)
def finished_experiment(shared_db_engine, shared_project_storage):
"""A successfully-run experiment. Its database schemas and project storage can be queried.
Returns: (triage.experiments.SingleThreadedExperiment)
"""
populate_source_data(shared_db_engine)
base_config = sample_config()
with mock.patch("triage.util.conf.open", side_effect=open_side_effect) as mock_file:
experiment = SingleThreadedExperiment(
base_config,
db_engine=shared_db_engine,
project_path=shared_project_storage.project_path,
)
experiment.run()
return experiment
def main():
args = parse_args()
dburl = os.environ['DBURL']
hiv_engine = create_engine(dburl, pool_pre_ping=True)
with open(args.config_path) as f:
experiment_config = yaml.load(f)
experiment = SingleThreadedExperiment(config=experiment_config,
db_engine=hiv_engine,
project_path=args.project_path,
replace=False)
experiment.validate()
experiment.run()
def prepare_experiment(config):
with testing.postgresql.Postgresql() as postgresql:
db_engine = create_engine(postgresql.url())
populate_source_data(db_engine)
with TemporaryDirectory() as temp_dir:
with mock.patch(
"triage.util.conf.open", side_effect=open_side_effect
) as mock_file:
experiment = SingleThreadedExperiment(
config=config,
db_engine=db_engine,
project_path=os.path.join(temp_dir, "inspections"),
cleanup=False,
partial_run=True,
)
yield experiment
def test_experiment_tracker_exception(db_engine, project_path):
with mock.patch("triage.util.conf.open",
side_effect=open_side_effect) as mock_file:
experiment = SingleThreadedExperiment(
config=sample_config(),
db_engine=db_engine,
project_path=project_path,
)
# no source data means this should blow up
with pytest.raises(Exception):
experiment.run()
with scoped_session(db_engine) as session:
experiment_run = session.query(TriageRun).get(experiment.run_id)
assert experiment_run.current_status == TriageRunStatus.failed
assert isinstance(experiment_run.last_updated_time, datetime.datetime)
assert experiment_run.stacktrace
def experiment(self):
db_url = self.root.db_url
config = yaml.load(self.args.config)
db_engine = create_engine(db_url)
common_kwargs = {
'db_engine': db_engine,
'project_path': self.args.project_path,
'config': config,
'replace': self.args.replace,
}
if self.args.n_db_processes > 1 or self.args.n_processes > 1:
experiment = MultiCoreExperiment(
n_db_processes=self.args.n_db_processes,
n_processes=self.args.n_processes,
**common_kwargs
)
else:
experiment = SingleThreadedExperiment(**common_kwargs)
return experiment
def model_evaluator(shared_db_engine, shared_project_storage):
"""Returns an instantiated ModelEvaluator available at module scope"""
populate_source_data(shared_db_engine)
base_config = sample_config()
# We need to have an ensemble model to test ModelEvaluator correctly
# so we can't use the finished_experiment fixture"""
base_config['grid_config'] = {
'sklearn.ensemble.ExtraTreesClassifier': {
'n_estimators': [10],
'criterion': ['gini'],
'max_depth': [1],
'max_features': ['sqrt'],
'min_samples_split': [2],
}
}
SingleThreadedExperiment(
base_config,
db_engine=shared_db_engine,
project_path=shared_project_storage.project_path).run()
return ModelEvaluator(1, 1, shared_db_engine)
def model_evaluator(shared_db_engine, shared_project_storage):
"""Returns an instantiated ModelEvaluator available at module scope"""
populate_source_data(shared_db_engine)
base_config = sample_config()
# We need to have an ensemble model to test ModelEvaluator correctly
# so we can't use the finished_experiment fixture"""
base_config["grid_config"] = {
"sklearn.ensemble.ExtraTreesClassifier": {
"n_estimators": [10],
"criterion": ["gini"],
"max_depth": [1],
"max_features": ["sqrt"],
"min_samples_split": [2],
}
}
with mock.patch("triage.util.conf.open",
side_effect=open_side_effect) as mock_file:
SingleThreadedExperiment(
base_config,
db_engine=shared_db_engine,
project_path=shared_project_storage.project_path,
).run()
return ModelEvaluator(1, 1, shared_db_engine)
import os
import sqlalchemy
import yaml
from catwalk.storage import FSModelStorageEngine
from triage.experiments import SingleThreadedExperiment
food_db = os.environ.get('FOOD_DB_URL')
print(food_db)
with open('inspections-training.yaml') as f:
experiment_config = yaml.load(f)
experiment = SingleThreadedExperiment(
config=experiment_config,
db_engine=sqlalchemy.create_engine(food_db),
model_storage_class=FSModelStorageEngine,
project_path='./triage-generated')
experiment.run()