import sys
from utils import dict_product, iwt, generate_configs
with open("../src/MuJoCo.json") as f:
BASE_CONFIG = json.load(f)
PARAMS = {
"game": ["Walker2d-v2"],
"mode": ["ppo"],
"out_dir": ["lstm_ppo_walker/agents"],
"norm_rewards": ["returns"],
"initialization": ["orthogonal"],
"anneal_lr": [True],
"value_clipping": [True],
"ppo_lr_adam": [1e-3],
"val_lr": [3e-2],
"cpu": [True],
"advanced_logging": [False],
"save_iters": [20],
"train_steps": [976],
"robust_ppo_eps": [0.05], # used for attack
"history_length": [100],
"use_lstm_val": [True],
"adv_clip_eps": [0.4],
"adv_entropy_coeff": [0.0],
"adv_ppo_lr_adam": [1e-3],
"adv_val_lr": [1e-2],
}
generate_configs(BASE_CONFIG, PARAMS)
print('...Testing...')
test(config)
# _ = utils.save_results(config, train_acc_per_epoch, test_acc_per_epoch, test_only, save_test)
if __name__ == '__main__':
# with torch.autograd.profiler.profile(use_cuda=True) as prof:
parser = argparse.ArgumentParser()
parser.add_argument("--dataset", type=str, default="MNIST")
parser.add_argument("-hr_test", type=float, default=None)
parser.add_argument("-config", type=str, required=True)
parser.add_argument("-colab", action="store_true")
parser.add_argument("-jit", action="store_true",
default=False) # can't plot results computed with jit
args = parser.parse_args()
hr_test = args.hr_test
# assert hr_test in [None, "1", "2"], "hr_test must be 1, 2 or None!!!"
run_configs = utils.load_run_config(args.config)
configs = utils.generate_configs(
run_configs, hr_test, args.colab, args.config,
args.jit) # list of configurations (=dict) to be trained
time0 = time() # total run time
for i, conf in enumerate(configs):
print(f'\n[{i + 1}/{len(configs)}] {conf["save_dir"]}')
try:
utils.reset_seed(conf['random_seed'])
main(conf)
except Exception as e:
raise
print('End')
import json
import sys
from utils import dict_product, iwt, generate_configs
with open("../src/MuJoCo.json") as f:
BASE_CONFIG = json.load(f)
from walker_atla_ppo_lstm_best import PARAMS as params
PARAMS = {
"mode": ["adv_sa_ppo"],
"out_dir": ["robust_atla_ppo_lstm_walker/agents"],
"robust_ppo_eps": [0.05],
"robust_ppo_reg": [0.3],
"robust_ppo_eps_scheduler_opts": ["start=1,length=1831"],
"robust_ppo_beta": [1.0],
"robust_ppo_beta_scheduler_opts": ["same"], # Using the same scheduler as eps scheduler
"robust_ppo_detach_stdev": [False],
"robust_ppo_method": ["sgld"],
"robust_ppo_pgd_steps": [2],
"adv_clip_eps": [0.2, 0.4],
"adv_entropy_coeff": [3e-4],
"adv_ppo_lr_adam": [1e-2],
"adv_val_lr": [1e-2],
}
params.update(PARAMS)
generate_configs(BASE_CONFIG, params)
if not test_only:
print('...Training...')
train_acc_per_epoch = train(config)
print('...Testing...')
test_acc_per_epoch = test(config)
_ = utils.save_results(config, train_acc_per_epoch, test_acc_per_epoch, test_only, save_test)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--dataset", type=str, default="MNIST")
parser.add_argument("-test_only", action="store_true")
parser.add_argument("-save_test", action="store_true")
args = parser.parse_args()
test_only, save_test = args.test_only, args.save_test
run_configs = utils.load_run_config('run_config.json')
np.random.seed(run_configs['random_seed']) # allows reproducibility
torch.manual_seed(run_configs['random_seed']) # allows reproducibility
configs = utils.generate_configs(run_configs, test_only) # list of configurations (=dict) to be trained
time0 = time() # total run time
for i, conf in enumerate(configs):
print(f'[{i + 1}/{len(configs)}] {conf["save_dir"]}')
try:
main(conf)
except Exception as e:
raise
print('End')