dt = datetime.datetime.now()
timestr = '{}-{}-{}-{}'.format(dt.month, dt.day, dt.hour, dt.minute)
for dat, (n_samples, sz) in dat_size_dict.items():
root_cmd = 'python bnn_stein_f.py -dataset={} '.format(dat)
param_specs = {
'layers': ['10 10'],
'seed': list(range(1334, 1344)),
'lr': [4e-3],
'n_particles': [20],
('batch_size', 'n_epoch'): [[100, 500]],
'ptb_scale': [1],
'test_freq': [20]
}
if n_samples > 1000:
param_specs[('batch_size', 'n_epoch')] = [[1000, 3000]]
if dat in set(['protein_data']):
param_specs['layers'] = ['10 10']
param_specs['seed'] = [1333, 1334, 1335, 1336, 1337]
log_dir = logdir_fmt.format(timestr, dat)
tasks += runner.list_tasks(root_cmd, param_specs, source_dir, log_dir)
print('\n'.join([t.cmd for t in tasks]))
print(len(tasks))
r = runner.Runner(n_max_gpus=4, n_multiplex=2, n_max_retry=1)
r.run_tasks(tasks)
assert (status == runner.Status.CRASHED) == (
task.option_dict['x'] < 0 or task.option_dict['sleep_long'] is True)
ctr[status] += 1
slave_working_dir = os.path.dirname(os.path.abspath(__file__))
param_specs = {
('x', 'x1'): [0.5, 0.6, -0.3], # x and x1 will have same value
'y': ['foo', 'bar!'],
'sleep_long': runner.BooleanOpt(true_first=False, no_for_false=True)
}
log_dir = '/tmp/slave_logs/'
if os.path.exists(log_dir):
shutil.rmtree(log_dir)
tasks = runner.list_tasks('python3 slave.py',
param_specs,
slave_working_dir,
log_dir + 'prefix',
max_cpu_time=2,
post_kill_cmd='echo "KILLED"')
r = runner.Runner(n_max_gpus=1,
n_multiplex=4,
n_max_retry=-1,
on_task_finish=on_task_finish)
r.run_tasks(tasks)
print(ctr)
dt = datetime.datetime.now()
timestr = '{}-{}-{}-{}'.format(dt.month, dt.day, dt.hour, dt.minute)
for dat, (n_samples, sz) in dat_size_dict.items():
root_cmd = 'python bnn_stein_f.py -dataset={} -valid '.format(dat)
param_specs = {
'layers': ['50 50'],
'seed': list(range(1334, 1354)),
'lr': [4e-3],
'n_particles': [20],
('batch_size', 'n_epoch'): [[100, 1000]],
'ptb_scale': [1],
'test_freq': [20]
}
if n_samples > 1000:
param_specs[('batch_size', 'n_epoch')] = [[1000, 1000]]
if dat in set(['protein_data']):
param_specs['layers'] = ['100 100']
param_specs['seed'] = [1333, 1334, 1335, 1336, 1337]
log_dir = logdir_fmt.format(timestr, dat)
tasks += runner.list_tasks(root_cmd, param_specs, source_dir, log_dir)
print('\n'.join([t.cmd for t in tasks]))
print(len(tasks))
r = runner.Runner(n_max_gpus=3, n_multiplex=1, n_max_retry=1)
r.run_tasks(tasks)
import logging
import os
import shutil
import sys
logging.basicConfig(stream=sys.stderr,
level=logging.DEBUG,
format='%(filename)s:%(lineno)s %(levelname)s:%(message)s')
slave_working_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'../src')
param_specs = {
'seed': list(range(1234, 1239)),
('z_dims', 'eps_dims'): [8, 32],
'lr': [1e-3],
'n_iter': [100000],
'optimizer': ['rmsprop'],
'model': ['vae']
}
log_dir = os.path.expanduser('~/s-vae-tf/run/svae/vae-explicit/'.format(
__file__.split('.')[0]))
tasks = runner.list_tasks('python mnist.py -explicit -test ', param_specs,
slave_working_dir, log_dir + 'prefix')
print('\n'.join([t.cmd for t in tasks]))
r = runner.Runner(n_max_gpus=6, n_multiplex=3, n_max_retry=-1)
r.run_tasks(tasks)
format='%(filename)s:%(lineno)s %(levelname)s:%(message)s')
slave_dir = os.path.abspath('.')
root_cmd = 'python main.py --layers 100,100 --bandit {}'.format(args.bandit)
param_specs = {'-num_context': [40000], '-seed': list(range(1334, 1344))}
import datetime
dt = datetime.datetime.now()
timestr = '{}-{}-{}-{}'.format(dt.month, dt.day, dt.hour, dt.minute)
log_dir = os.path.join(slave_dir,
'../../run/bandit-{}-{}/'.format(args.bandit, timestr))
os.makedirs(os.path.join(log_dir, "test_{}".format(
dt.second))) # make sure we can write
tasks = runner.list_tasks(root_cmd, param_specs, slave_dir, log_dir)
for i, t in enumerate(tasks):
ncmd = t.cmd.replace('-dir', '--logdir').replace('-production ', '')
tasks[i] = t._replace(cmd=ncmd)
print('\n'.join([t.cmd for t in tasks]))
print(args)
# r = runner.Runner(n_max_gpus=[0,3], n_multiplex=5, n_max_retry=-1)
# r = runner.Runner(n_max_gpus=7, n_multiplex=2, n_max_retry=-1)
r = runner.Runner(n_max_gpus=args.max_gpus,
n_multiplex=args.n_multiplex,
n_max_retry=-1)
r.run_tasks(tasks)