diff --git a/rlkit/launchers/launcher_util.py b/rlkit/launchers/launcher_util.py index d50eb71b9..87cb7be55 100644 --- a/rlkit/launchers/launcher_util.py +++ b/rlkit/launchers/launcher_util.py @@ -90,7 +90,7 @@ def run_experiment_here( variant=None, exp_id=0, seed=None, - use_gpu=True, + use_gpu=False, # Logger params: exp_prefix="default", snapshot_mode='last', @@ -292,7 +292,11 @@ def setup_logger( f.write(code_diff + '\n') if code_diff_staged is not None and len(code_diff_staged) > 0: with open(osp.join(log_dir, diff_staged_file_name), "w") as f: - f.write(code_diff_staged + '\n') + try: + f.write(code_diff_staged + '\n') + except UnicodeEncodeError as e: + print(e) + f.write(code_diff_staged.encode('utf-8', 'surrogateescape').decode('ISO-8859-1') + '\n') with open(osp.join(log_dir, "git_infos.txt"), "a") as f: f.write("directory: {}\n".format(directory)) f.write("git hash: {}\n".format(commit_hash)) @@ -565,7 +569,7 @@ def foo(variant): )) except git.exc.InvalidGitRepositoryError: pass - except ImportError: + except (ImportError, UnboundLocalError): git_infos = None run_experiment_kwargs = dict( exp_prefix=exp_prefix, diff --git a/scripts/sim_goal_conditioned_policy.py b/scripts/sim_goal_conditioned_policy.py deleted file mode 100644 index 70a48ca35..000000000 --- a/scripts/sim_goal_conditioned_policy.py +++ /dev/null @@ -1,62 +0,0 @@ -import argparse -import pickle - -from rlkit.core import logger -from rlkit.samplers.rollout_functions import multitask_rollout -from rlkit.torch import pytorch_util as ptu -from rlkit.envs.vae_wrapper import VAEWrappedEnv - - -def simulate_policy(args): - if args.pause: - import ipdb; ipdb.set_trace() - data = pickle.load(open(args.file, "rb")) - policy = data['policy'] - env = data['env'] - print("Policy and environment loaded") - if args.gpu: - ptu.set_gpu_mode(True) - policy.to(ptu.device) - if isinstance(env, VAEWrappedEnv): - env.mode(args.mode) - if args.enable_render or hasattr(env, 'enable_render'): - # some environments need to be reconfigured for visualization - env.enable_render() - policy.train(False) - paths = [] - while True: - paths.append(multitask_rollout( - env, - policy, - max_path_length=args.H, - animated=not args.hide, - observation_key='observation', - desired_goal_key='desired_goal', - )) - if hasattr(env, "log_diagnostics"): - env.log_diagnostics(paths) - if hasattr(env, "get_diagnostics"): - for k, v in env.get_diagnostics(paths).items(): - logger.record_tabular(k, v) - logger.dump_tabular() - - -if __name__ == "__main__": - - parser = argparse.ArgumentParser() - parser.add_argument('file', type=str, - help='path to the snapshot file') - parser.add_argument('--H', type=int, default=300, - help='Max length of rollout') - parser.add_argument('--speedup', type=float, default=10, - help='Speedup') - parser.add_argument('--mode', default='video_env', type=str, - help='env mode') - parser.add_argument('--gpu', action='store_true') - parser.add_argument('--pause', action='store_true') - parser.add_argument('--enable_render', action='store_true') - parser.add_argument('--multitaskpause', action='store_true') - parser.add_argument('--hide', action='store_true') - args = parser.parse_args() - - simulate_policy(args) diff --git a/scripts/sim_policy.py b/scripts/sim_policy.py deleted file mode 100644 index 89baf6f68..000000000 --- a/scripts/sim_policy.py +++ /dev/null @@ -1,43 +0,0 @@ -from rlkit.samplers.util import rollout -from rlkit.torch.core import PyTorchModule -from rlkit.torch.pytorch_util import set_gpu_mode -import argparse -import joblib -import uuid -from rlkit.core import logger - -filename = str(uuid.uuid4()) - - -def simulate_policy(args): - data = joblib.load(args.file) - policy = data['policy'] - env = data['env'] - print("Policy loaded") - if args.gpu: - set_gpu_mode(True) - policy.cuda() - if isinstance(policy, PyTorchModule): - policy.train(False) - while True: - path = rollout( - env, - policy, - max_path_length=args.H, - animated=True, - ) - if hasattr(env, "log_diagnostics"): - env.log_diagnostics([path]) - logger.dump_tabular() - - -if __name__ == "__main__": - parser = argparse.ArgumentParser() - parser.add_argument('file', type=str, - help='path to the snapshot file') - parser.add_argument('--H', type=int, default=300, - help='Max length of rollout') - parser.add_argument('--gpu', action='store_true') - args = parser.parse_args() - - simulate_policy(args) diff --git a/scripts/sim_tdm_policy.py b/scripts/sim_tdm_policy.py deleted file mode 100644 index d7b2272a9..000000000 --- a/scripts/sim_tdm_policy.py +++ /dev/null @@ -1,64 +0,0 @@ -import argparse -import json - -import joblib -from pathlib import Path - -import rlkit.torch.pytorch_util as ptu -from rlkit.core.eval_util import get_generic_path_information -from rlkit.torch.tdm.sampling import multitask_rollout -from rlkit.core import logger -if __name__ == "__main__": - - parser = argparse.ArgumentParser() - parser.add_argument('file', type=str, help='path to the snapshot file') - parser.add_argument('--H', type=int, default=300, - help='Max length of rollout') - parser.add_argument('--nrolls', type=int, default=1, - help='Number of rollout per eval') - parser.add_argument('--mtau', type=float, help='Max tau value') - parser.add_argument('--gpu', action='store_true') - parser.add_argument('--hide', action='store_true') - args = parser.parse_args() - - data = joblib.load(args.file) - if args.mtau is None: - # Load max tau from variant.json file - variant_path = Path(args.file).parents[0] / 'variant.json' - variant = json.load(variant_path.open()) - try: - max_tau = variant['tdm_kwargs']['max_tau'] - print("Max tau read from variant: {}".format(max_tau)) - except KeyError: - print("Defaulting max tau to 0.") - max_tau = 0 - else: - max_tau = args.mtau - - env = data['env'] - policy = data['policy'] - policy.train(False) - - if args.gpu: - ptu.set_gpu_mode(True) - policy.cuda() - - while True: - paths = [] - for _ in range(args.nrolls): - goal = env.sample_goal_for_rollout() - path = multitask_rollout( - env, - policy, - init_tau=max_tau, - goal=goal, - max_path_length=args.H, - animated=not args.hide, - cycle_tau=True, - decrement_tau=True, - ) - paths.append(path) - env.log_diagnostics(paths) - for key, value in get_generic_path_information(paths).items(): - logger.record_tabular(key, value) - logger.dump_tabular()