def main():
parser = argparse.ArgumentParser()
parser.add_argument('-e', '--exp_name', default="acacia")
parser.add_argument('-m', '--mode', default=None)
parser.add_argument('-image_path', '--image_path', default=None)
parser.add_argument('-model_path', '--model_path', default=None)
parser.add_argument('-model_name', '--model_name', default=None)
parser.add_argument(
'-r',
'--reset',
action="store_const",
const=True,
default=False,
help="If set, a new model will be created, overwriting any previous version.")
args = parser.parse_args()
dataset_name, model_name, metric_name = experiments.get_experiment(
args.exp_name)
# Paths
name = "{}_{}".format(dataset_name, model_name)
# Train checkpoint file
path_model = "checkpoints/best_model_{}.pth".format(name)
path_opt = "checkpoints/opt_{}.pth".format(name)
path_best_model = "checkpoints/best_model_save_{}.pth".format(name)
path_history = "checkpoints/history_{}.json".format(name)
if args.image_path is not None:
applyOnImage.apply(args.image_path, args.model_name, args.model_path)
elif args.mode == "train":
train.train(
dataset_name,
model_name,
metric_name,
path_history,
path_model,
path_opt,
path_best_model,
args.reset)
elif args.mode == "test":
test.test(
dataset_name,
model_name,
metric_name,
path_history,
path_model)
elif args.mode == "summary":
summary.summary(dataset_name, model_name, path_history)
def __init__(self, env_seed, env_name='', deltas=None, delta_std=0.02):
self.env = gym.make(env_name)
self.env.seed(0)
self.params = {}
self.params['env_name'] = env_name
self.params['ob_dim'] = self.env.observation_space.shape[0]
self.params['ac_dim'] = self.env.action_space.shape[0]
self.params = get_experiment(self.params)
self.params['zeros'] = True
self.params['seed'] = 0
self.deltas = SharedNoiseTable(deltas, env_seed + 7)
self.delta_std = delta_std
def main():
parser = argparse.ArgumentParser()
## Env setup
parser.add_argument('--env_name', type=str, default='point-v0')
parser.add_argument('--num_agents', '-na', type=int, default=5)
parser.add_argument('--seed', '-sd', type=int, default=0)
parser.add_argument('--max_iter', '-it', type=int, default=2000)
parser.add_argument('--policy', '-po', type=str, default='FC')
parser.add_argument('--embedding', '-em', type=str, default='a_s')
parser.add_argument('--num_workers', '-nw', type=int, default=4)
parser.add_argument('--filename', '-f', type=str, default='')
parser.add_argument('--num_evals', '-ne', type=int, default=0)
parser.add_argument('--flush', '-fl', type=int, default=1000) # may need this. it resets ray, because sometimes it fills the memory.
parser.add_argument('--ob_filter', '-ob', type=str, default='MeanStdFilter') # 'NoFilter'
parser.add_argument('--w_nov', '-wn', type=float, default=-1) # if negative it uses the adaptive method, else it will be fixed at the value you pick (0,1). Note that if you pick 1 itll be unsupervised (ie no reward)
parser.add_argument('--dpp_kernel', '-ke', type=str, default='rbf')
parser.add_argument('--states', '-ss', type=str, default='random-20') # 'random-X' X is how many
parser.add_argument('--update_states', '-us', type=int, default=20) # how often to update.. we only used 20
args = parser.parse_args()
params = vars(args)
params = get_experiment(params)
ray.init()
os.environ['RAY_USE_NEW_GCS'] = 'True'
state_word = [str(params['states'].split('-')[0]) if params['w_nov'] > 0 else ''][0]
params['dir'] = params['env_name'] + '_Net' + str(params['layers']) + 'x' + str(params['h_dim']) + '_Agents' + str(params['num_agents']) + '_Novelty' + str(params['w_nov']) + state_word + 'kernel_' + params['dpp_kernel'] + '_lr' + str(params['learning_rate']) + '_' + params['filename'] + params['ob_filter']
if not(os.path.exists('data/'+params['dir'])):
os.makedirs('data/'+params['dir'])
os.makedirs('data/'+params['dir']+'/weights')
os.makedirs('data/'+params['dir']+'/results')
train(params)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--env_name', type=str, default='Swimmer-v2')
parser.add_argument('--seed', '-sd', type=int, default=0)
parser.add_argument('--max_iter', '-it', type=int, default=200)
parser.add_argument('--num_workers', '-nw', type=int, default=4)
parser.add_argument('--filename', '-f', type=str, default='') # anything else you want to add
args = parser.parse_args()
params = vars(args)
params = get_experiment(params)
params['dir'] = params['env_name'] + '_' + params['filename']
if not(os.path.exists('data/'+params['dir'])):
os.makedirs('data/'+params['dir'])
os.makedirs('data/'+params['dir']+'/results')
ray.init()
train(params)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('-e', '--exp')
parser.add_argument('-m', '--mode')
parser.add_argument('-r', '--reset', default=0, type=int)
parser.add_argument('-c', '--cut', default=None, type=int)
parser.add_argument('-dList', '--dList', default=None, type=str, nargs="+")
parser.add_argument('-mList', '--mList', default=None, type=str, nargs="+")
parser.add_argument('-eList', '--eList', default=None, type=int, nargs="+")
parser.add_argument('-lList', '--lList', default=None, nargs="+")
parser.add_argument('-sList', '--sList', default=None, type=str, nargs="+")
args = parser.parse_args()
dList, mList, eList, lList, sList = experiments.get_experiment(
args.exp, args)
mList.sort()
sList.sort()
create_plot = False
results = {}
for d, m, e, l, s in product(dList, mList, eList, lList, sList):
history = ut.load_history(d, m, l, e, s, args.reset)
if args.mode == "train":
if len(history["loss"]) == 0 or args.reset:
train.train(dataset_name=d,
model_name=m,
epochs=e,
learning_rate=l,
sampling_method=s,
def main():
# SET SEED
np.random.seed(1)
torch.manual_seed(1)
torch.cuda.manual_seed_all(1)
# SEE IF CUDA IS AVAILABLE
#print("STARTING - CUDA: %s" % torch.cuda.is_available())
assert torch.cuda.is_available()
print("CUDA: %s" % torch.version.cuda)
print("Pytroch: %s" % torch.__version__)
parser = argparse.ArgumentParser()
parser.add_argument('-e', '--exp')
parser.add_argument('-b', '--borgy', default=0, type=int)
parser.add_argument('-br', '--borgy_running', default=0, type=int)
parser.add_argument('-m', '--mode', default="summary")
parser.add_argument('-r', '--reset', default="None")
parser.add_argument('-g', '--gpu', type=int)
parser.add_argument('-c', '--configList', nargs="+", default=None)
parser.add_argument('-l', '--lossList', nargs="+", default=None)
parser.add_argument('-d', '--datasetList', nargs="+", default=None)
parser.add_argument('-metric', '--metricList', nargs="+", default=None)
args = parser.parse_args()
mu.set_gpu(args.gpu)
mode = args.mode
exp_dict = experiments.get_experiment(args.exp)
if args.configList is None:
configList = exp_dict["configList"]
else:
configList = args.configList
if args.metricList is None:
metricList = exp_dict["metricList"]
else:
metricList = args.metricList
if args.datasetList is None:
datasetList = exp_dict["datasetList"]
else:
datasetList = args.datasetList
epochs = exp_dict["epochs"]
if args.lossList is None:
lossList = exp_dict["lossList"]
else:
lossList = args.lossList
# vis.close(env="loss")
# vis.close(env="plots")
results = {}
for config, metric, dataset, loss_name in product(configList, metricList,
datasetList, lossList):
argList = mu.get_argList(mode, dataset, config, args.reset, epochs,
metric, loss_name)
main_dict = mu.argList2main_dict(main_parser.get_parser(), argList)
key = ("%s - %s" % (config, loss_name), "%s_(%s)" % (dataset, metric))
if mode == "summary":
results[key] = mu.summary(argList, main_dict, which="train")
# TRAIN CASES
if mode == "debug":
debug.debug(main_dict)
if mode == "train":
train.main(main_dict)
# TEST CASES
if mode == "test":
print(main_dict["dataset_name"])
cm = test.test(main_dict)
print(cm)
print(ut.dict2frame(results))