1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193 | """Argument parser and logic for the ``record-video`` sub-command."""
from __future__ import annotations
import argparse
import os
from pathlib import Path
import sys
from typing import Any
import numpy as np
import yaml
from lit_sb3.core.exp_manager import ExperimentManager
from lit_sb3.huggingface.naming_schemes import EnvironmentName
from lit_sb3.utils.cli import StoreDict
from lit_sb3.utils.env import create_test_env
from lit_sb3.utils.model import get_model_path, get_saved_hyperparams
from lit_sb3.utils.registry import ALGOS
from lit_sb3.utils.training import set_random_seed
from lit_sb3.vec_env import VecVideoRecorder
def _add_record_video_args(parser: argparse.ArgumentParser) -> None:
"""Add record-video arguments to *parser*."""
parser.add_argument("--env", help="Environment ID", type=str, default="CartPole-v1")
parser.add_argument("-f", "--folder", help="Log folder", type=str, default="rl-trained-agents")
parser.add_argument("-o", "--output-folder", help="Output folder", type=str)
parser.add_argument(
"--algo",
help="RL Algorithm",
default="ppo",
type=str,
required=False,
choices=list(ALGOS.keys()),
)
parser.add_argument("-n", "--n-timesteps", help="Number of timesteps", default=1000, type=int)
parser.add_argument("--n-envs", help="Number of environments", default=1, type=int)
parser.add_argument("--deterministic", action="store_true", default=False, help="Use deterministic actions")
parser.add_argument("--stochastic", action="store_true", default=False, help="Use stochastic actions")
parser.add_argument("--seed", help="Random generator seed", type=int, default=0)
parser.add_argument(
"--no-render",
action="store_true",
default=False,
help="Do not render the environment (useful for tests)",
)
parser.add_argument(
"--exp-id",
help="Experiment ID (default: 0: latest, -1: no exp folder)",
default=0,
type=int,
)
parser.add_argument(
"--load-best",
action="store_true",
default=False,
help="Load best model instead of last model if available",
)
parser.add_argument(
"--load-checkpoint",
type=int,
help="Load checkpoint instead of last model if available, you must pass the number of timesteps corresponding to it",
)
parser.add_argument(
"--load-last-checkpoint",
action="store_true",
default=False,
help="Load last checkpoint instead of last model if available",
)
parser.add_argument(
"--env-kwargs",
type=str,
nargs="+",
action=StoreDict,
help="Optional keyword argument to pass to the env constructor",
)
parser.add_argument(
"--custom-objects",
action="store_true",
default=False,
help="Use custom objects to solve loading issues",
)
def run_record_video(args: argparse.Namespace) -> None: # noqa: C901
"""Execute record-video from a parsed :class:`argparse.Namespace`."""
import lit_sb3.core.import_envs # noqa: F401
env_name: EnvironmentName = EnvironmentName(args.env)
algo = args.algo
folder = args.folder
video_folder = args.output_folder
seed = args.seed
video_length = args.n_timesteps
n_envs = args.n_envs
name_prefix, model_path, log_path = get_model_path(
args.exp_id,
folder,
algo,
env_name,
args.load_best,
args.load_checkpoint,
args.load_last_checkpoint,
)
print(f"Loading {model_path}")
off_policy_algos = ["qrdqn", "dqn", "ddpg", "sac", "her", "td3", "tqc", "crossq"]
set_random_seed(args.seed)
is_atari = ExperimentManager.is_atari(env_name.gym_id)
is_minigrid = ExperimentManager.is_minigrid(env_name.gym_id)
stats_path = Path(log_path) / str(env_name)
hyperparams, maybe_stats_path = get_saved_hyperparams(stats_path)
env_kwargs: dict[str, Any] = {}
args_path = Path(log_path) / str(env_name) / "args.yml"
if Path(args_path).is_file():
with Path(args_path).open() as f:
loaded_args = yaml.safe_load(f)
if loaded_args["env_kwargs"] is not None:
env_kwargs = loaded_args["env_kwargs"]
if args.env_kwargs is not None:
env_kwargs.update(args.env_kwargs)
env_kwargs.update(render_mode="rgb_array")
env = create_test_env(
env_name.gym_id,
n_envs=n_envs,
stats_path=maybe_stats_path,
seed=seed,
log_dir=None,
should_render=not args.no_render,
hyperparams=hyperparams,
env_kwargs=env_kwargs,
)
kwargs: dict[str, Any] = {"seed": args.seed}
if algo in off_policy_algos:
kwargs["buffer_size"] = 1
if "optimize_memory_usage" in hyperparams:
kwargs["optimize_memory_usage"] = False
newer_python_version = sys.version_info >= (3, 8)
custom_objects: dict[str, Any] = {}
if newer_python_version or args.custom_objects:
custom_objects = {
"learning_rate": 0.0,
"lr_schedule": lambda _: 0.0,
"clip_range": lambda _: 0.0,
}
model = ALGOS[algo].load(model_path, env=env, custom_objects=custom_objects, **kwargs)
stochastic = args.stochastic or ((is_atari or is_minigrid) and not args.deterministic)
deterministic = not stochastic
if video_folder is None:
video_folder = Path(log_path) / "videos"
env = VecVideoRecorder(
env,
video_folder,
record_video_trigger=lambda x: x == 0,
video_length=video_length,
name_prefix=name_prefix,
)
obs = env.reset()
lstm_states = None
episode_starts = np.ones((env.num_envs,), dtype=bool)
try:
for _ in range(video_length):
action, lstm_states = model.predict(
obs, # type: ignore[arg-type]
state=lstm_states,
episode_start=episode_starts,
deterministic=deterministic,
)
if not args.no_render:
env.render()
obs, _, dones, _ = env.step(action)
episode_starts = dones
except KeyboardInterrupt:
pass
env.close()
|