def test_init(self):
data = BasicConfig(**self.d)
self.assertDictEqual(data, self.d)
data = BasicConfig(a=1)
self.assertDictEqual(data, self.d)
self.assertRaises(ValueError, data.load, "wrong.data")
class CPUResourceManagerTestCase(unittest.TestCase):
n_parallel = 1
auppath = os.path.join("tests", "data", ".aup")
job = Job("test_Job.py", BasicConfig(), "./tests/EE")
job_fail = Job("task1.py", BasicConfig(), "./tests/data")
job_fail2 = Job("task2.py", BasicConfig(), "./tests/data")
job_fail3 = Job("task3.py", BasicConfig(), "./tests/data")
ori_db = os.path.join(auppath, "sqlite3.db")
bk_db = os.path.join(auppath, "bk.db")
val = 0
def setUp(self):
copyfile(self.ori_db, self.bk_db)
self.connector = get_default_connector(self.auppath)
self.rm = CPUResourceManager(self.connector, self.n_parallel)
def tearDown(self):
copyfile(self.bk_db, self.ori_db)
os.remove(self.bk_db)
def test_get_resource(self):
self.assertIsInstance(self.rm.get_available("test", "cpu"), int)
# !IMPORTANT - currently, there is no user authentication.
self.assertIsInstance(self.rm.get_available("test2", "cpu"), int)
def test_run(self):
def callback(*args):
self.val = -1
self.rm.run(self.job, self.rm.get_available("test", "cpu"), {},
callback)
self.rm.executor.shutdown(wait=True)
self.assertEqual(self.val, -1)
def test_fail_run(self):
def callback(*args):
raise ChildProcessError
self.rm.run(self.job_fail, self.rm.get_available("test", "cpu"), {},
callback)
self.rm.run(self.job_fail2, self.rm.get_available("test", "cpu"), {},
callback)
self.rm.run(self.job_fail3, self.rm.get_available("test", "cpu"), {},
callback)
def test_full_run(self):
def callback(*args):
pass
self.rm.run_job(self.job, self.rm.get_available("test", "cpu"), {},
callback)
self.rm.finish_job(self.job.jid, 2.0)
self.rm.executor.shutdown(wait=True)
self.assertListEqual([1, 0.1], self.rm.finish(1))
def test_finish(self):
self.rm.suspend()
self.assertListEqual([1, 0.1], self.rm.finish(1))
def test_pkl(self):
data = BasicConfig()
data.load(self.pkl_read)
self.assertDictEqual(self.d, data)
data["b"] = 2
data.c = 3
data.save(self.pkl_write)
data = BasicConfig().load(self.pkl_write)
self.assertEqual(data.b, 2)
self.assertEqual(data["c"], 3)
class GPUResourceManagerTestCase(unittest.TestCase):
n_parallel = 1
auppath = os.path.join("tests", "data", ".aup")
connector = get_default_connector(auppath)
val = 0
ori_db = os.path.join(auppath, "sqlite3.db")
bk_db = os.path.join(auppath, "bk.db")
job = Job("test_Job.py", BasicConfig(), "./tests/EE")
def setUp(self):
copyfile(self.ori_db, self.bk_db)
self.connector = get_default_connector(self.auppath)
self.rm = GPUResourceManager(self.connector,
self.n_parallel,
auppath=self.auppath)
def tearDown(self):
copyfile(self.bk_db, self.ori_db)
os.remove(self.bk_db)
def test_run(self):
def callback(*args):
self.val = -1
self.rm.run(self.job, self.rm.get_available("test", "gpu"), {},
callback)
sleep(3) # wait till subprocess finished - handled by rm.finish()
self.assertEqual(self.val, -1)
def test_run_through(self):
exp = Experiment(BasicConfig().load(self.path),
username="******", auppath=self.auppath)
exp.start()
exp.finish()
self.assertDictEqual(exp.pending_jobs, {})
def main(unused):
## change
config = BasicConfig().load(unused[1])
config.to_flags(FLAGS)
##
val = rosenbrock(FLAGS.x, FLAGS.y)
print_result(val)
def test_job_retries(self):
exp = Experiment(BasicConfig().load(
os.path.join("tests", "data", "exp6.json")),
username="******",
auppath=os.path.join("tests", "data", ".aup"))
exp.start()
jid, val = exp.finish()
self.assertDictEqual(exp.pending_jobs, {})
self.assertEqual(jid, 10)
class PassiveRMTestCase(unittest.TestCase):
auppath = os.path.join("tests", "data", ".aup")
job = Job("task1.py", BasicConfig(), "./tests/EE")
ori_db = os.path.join(auppath, "sqlite3.db")
bk_db = os.path.join(auppath, "bk.db")
val = 0
def setUp(self):
copyfile(self.ori_db, self.bk_db)
self.connector = get_default_connector(self.auppath)
self.rm = PassiveResourceManager(self.connector)
def tearDown(self):
copyfile(self.bk_db, self.ori_db)
os.remove(self.bk_db)
def test_get_available(self):
self.assertEqual(10, self.rm.get_available("test", "passive"))
self.connector.take_available_resource(10)
self.assertEqual(None, self.rm.get_available("test", "passive"))
self.rm.running = True
self.assertEqual(None, self.rm.get_available("test", "passive"))
#!/usr/bin/env python3
"""
Job will fail when sleep time is 1
============================================
..
Copyright (c) 2018 LG Electronics Inc.
SPDX-License-Identifier: GPL-3.0-or-later
"""
import sys
from time import sleep
from aup import BasicConfig, print_result
if __name__ == "__main__":
if len(sys.argv) != 2:
print("config file required")
exit(1)
config = BasicConfig().load(sys.argv[1])
sleep(1 + config.time / 10.)
if config.time == 3:
exit(1)
print_result(config.time)
def test_ssh_session(self):
pkey = pkg_resources.resource_filename("mockssh", 'server-key')
data_folder = pkg_resources.resource_filename(__name__, 'data')
users = {
"mockuser": pkey,
}
scores = {}
mutex = Lock()
config = {
'workingdir':
data_folder,
'n_parallel':
2,
'resource':
'node',
'proposer':
'random',
'script':
os.path.join(data_folder, "script", "rosenbrock_hpo.py"),
'n_samples':
10,
'random_seed':
10,
'parameter_config': [{
'name': 'x',
'type': 'float',
'range': [-5, 5]
}, {
'name': 'y',
'type': 'float',
'range': [-5, 5]
}],
"runtime_args": {
"prescript": "export CUDA_VISIBLE_DEVICES=-1",
"postscript": "echo $CUDA_VISIBLE_DEVICES",
"overwrite": 'true'
}
}
m = rp.RandomProposer(config)
def callback_fun(score, jid):
mutex.acquire()
scores[jid] = score
mutex.release()
with mockssh.Server(users) as server1, mockssh.Server(
users) as server2:
mngr = SSHResourceManager(None,
config['n_parallel'],
auppath=data_folder,
reconn_wait_time=3,
max_retries=3)
mngr.mapping[1] = "[email protected]:" + str(
server1.port) + " " + pkey
mngr.mapping[2] = "[email protected]:" + str(
server2.port) + " " + pkey
for i in range(0, config['n_samples'], 2):
proposal = m.get()
job1 = Job(config['script'], BasicConfig(**proposal),
config['workingdir'])
job1.jid = i
mngr.run(job1, 1, None, callback_fun, overwrite=True)
proposal = m.get()
job2 = Job(config['script'], BasicConfig(**proposal),
config['workingdir'])
job2.jid = i + 1
mngr.run(job2, 2, None, callback_fun, overwrite=True)
mngr.executor.shutdown(wait=True)
for key in scores:
self.assertTrue(scores[key] != "ERROR")
print("config file required")
exit(1)
parser = argparse.ArgumentParser()
parser.add_argument('--dropout',
type=float,
default=0.2,
help='drop probability for training dropout.')
parser.add_argument('--textlstm',
type=int,
default=256,
help='text LSTM layer size.')
parser.add_argument('--speechlstm',
type=int,
default=128,
help='speech layer size.')
parser.add_argument('--finalfc',
type=int,
default=256,
help='final FC layer size.')
parser.add_argument('--adamlr', type=float, default=0.001, help='adam lr')
FLAGS, unparsed = parser.parse_known_args()
config = BasicConfig(**FLAGS.__dict__).load(sys.argv[1])
FLAGS.__dict__.update(config)
val = train()
print(str(val))
print_result(val)
def main():
print("Starting Client")
config = BasicConfig().load(sys.argv[1])
run(config["expdir"])
def test_hash(self):
a = {"a": 1}
b = BasicConfig(a=1)
self.assertDictEqual(a, b)
def test_type(self):
data = BasicConfig()
self.assertRaises(TypeError, data.load, self.wrong_read)
def test_compare(self):
data = BasicConfig(**self.d)
self.assertTrue(data == self.d)
def test_access(self):
data = BasicConfig(**self.d)
del data.a
self.assertFalse('a' in data)