def mp_run(cfg):
print('mp_run of {}'.format(cfg))
bailing = None
# Try to read configuration
file = None
try:
file = open(cfg, 'r')
except Exception as e:
bailing = "mp_run {}:\n{}\n".format(cfg, traceback.format_exc())
# Run the experiment
if file:
with file:
args = yaml.load(file)
try:
cfg_run(**args)
except Exception as e:
bailing = "mp_run {}:\n{}\n".format(cfg, traceback.format_exc())
# take care of fails
if bailing:
f = open("bailing.out", "a")
try:
f.write(bailing + "\n")
finally:
f.close()
def mp_run(cfg):
# Multiple copies can be run on one computer at the same time, which results in the same seed for a random generator.
# Thus we need to wait for a second or so between runs
global counter
global cores
with counter_lock:
wait = counter.value
counter.value += 2
sleep(wait)
print('wait finished {0}'.format(wait))
# Run the experiment
with open(cfg, 'r') as file:
args = yaml.load(file)
cfg_run(**args)
def mp_run(cfg):
print('mp_run of {}'.format(cfg))
# Read configuration
try:
file = open(cfg, 'r')
except IOError:
print("Could not read file: {}".format(cfg))
sys.exit()
with file:
args = yaml.load(file)
# Run the experiment
try:
cfg_run(**args)
except Exception:
print('mp_run {} failid to exit correctly'.format(cfg))
sys.exit()
def mujoco_export(env,
mp,
task='Walking',
policies='',
trajectories='',
misc=''):
args = parse_args()
if task == 'Balancing':
task_balancing = task
else:
task_balancing = ''
args['cfg'] = "Roboschool{}-v1".format(env + task_balancing + 'GRL')
args['steps'] = 0
args['trials'] = 1
args['test_interval'] = 0
args['normalize_observations'] = False
args['normalize_returns'] = False
args['batch_norm'] = True
args['render'] = True
args['output'] = misc + '{}_{}_play-mp{}'.format(env.lower(), task.lower(),
mp)
t = task[0].lower()
if t == 'b':
stage = stage_names[1]
elif t == 'w':
stage = stage_names[2]
else:
raise ValueError('incorrect task ' + task)
args[
'load_file'] = policies + 'ddpg-exp1_two_stage_{env}_ga_{task}-g0001-mp{mp}-{stage}'.format(
env=env.lower(), task=t, mp=mp, stage=stage)
args[
'compare_with'] = policies + 'ddpg-exp1_two_stage_{env}_ga_b-g0001-mp{mp}-01_balancing-last'.format(
env=env.lower(), mp=mp)
args['trajectory'] = trajectories + '{}_{}-mp{}'.format(
env.lower(), task.lower(), mp)
args['env_timestep'] = 0.0165
# Run actual script.
args['save'] = False
cfg_run(**args)
def leo_export(mp, policies='', task='walking', trajectories='', misc=''):
args = parse_args()
env = 'leo'
args['cfg'] = 'cfg/{}_{}_play.yaml'.format(env, task)
args['steps'] = 0
args['trials'] = 1
args['test_interval'] = 0
args['normalize_observations'] = False
args['normalize_returns'] = False
args['batch_norm'] = True
args['output'] = misc + '{}_{}_play-mp{}'.format(env, task, mp)
t = task[0].lower()
if t == 'b':
stage = stage_names[1]
elif t == 'w':
stage = stage_names[2]
else:
raise ValueError('incorrect task ' + task)
args[
'load_file'] = policies + 'ddpg-exp1_two_stage_leo_ga_{task}-g0001-mp{mp}-{stage}'.format(
task=t, mp=mp, stage=stage)
args[
'compare_with'] = policies + 'ddpg-exp1_two_stage_leo_ga_b-g0001-mp{mp}-01_balancing-last'.format(
mp=mp)
args['trajectory'] = trajectories + '{}_{}-mp{}'.format(env, task, mp)
args['env_timestep'] = 0.03
# Run actual script.
args['save'] = False
cfg_run(**args)
if task == 'walking':
os.rename('aux_leo.csv', 'aux_leo-mp{}.csv'.format(mp))
task_balancing = task
else:
task_balancing = ''
if tf:
tfstr = '_TF'
else:
tfstr = ''
args['cfg'] = "Roboschool{}-v1".format(env + task_balancing + 'GRL' + tfstr)
#args['cfg'] = "Roboschool{}-v1".format(env+task_balancing)
#args['cfg'] = "Roboschool{}-v1".format(env+task)
args['steps'] = 10000
args['test_interval'] = 30
#args['seed'] = 1
args['rb_max_size'] = args['steps']
args['normalize_observations'] = False
args['normalize_returns'] = False
args['batch_norm'] = True
args['version'] = 0
args['output'] = 'cl/{}_{}{}'.format(env.lower(), task.lower(), tfstr.lower())
args['rb_save_filename'] = 'cl/{}_{}{}'.format(env.lower(), task.lower(),
tfstr.lower())
#args['rb_load_filename'] = '{}_{}'.format(env, 'balancing')
#args['reassess_for'] = 'walking_1_0'
# Run actual script.
args['save'] = True
cfg_run(**args)