def build_and_train(slot_affinity_code, log_dir, run_ID, config_key):
affinity = affinity_from_code(slot_affinity_code)
config = configs[config_key]
variant = load_variant(log_dir)
config = update_config(config, variant)
sampler = AsyncSerialSampler(
EnvCls=gym_make,
env_kwargs=config["env"],
eval_env_kwargs=config["env"],
CollectorCls=DbCpuResetCollector,
**config["sampler"]
)
algo = DDPG(optim_kwargs=config["optim"], **config["algo"])
agent = DdpgAgent(**config["agent"])
runner = AsyncRlEval(
algo=algo,
agent=agent,
sampler=sampler,
affinity=affinity,
**config["runner"]
)
name = "async_ddpg_" + config["env"]["id"]
with logger_context(log_dir, run_ID, name, config):
runner.train()
def build_and_train(game="pong", run_ID=0):
# Change these inputs to match local machine and desired parallelism.
affinity = make_affinity(
run_slot=0,
n_cpu_core=2, # Use 16 cores across all experiments.
n_gpu=1, # Use 8 gpus across all experiments.
sample_gpu_per_run=0,
async_sample=True, # Different affinity structure fo async.
# hyperthread_offset=24, # If machine has 24 cores.
# n_socket=2, # Presume CPU socket affinity to lower/upper half GPUs.
# gpu_per_run=2, # How many optimizer GPUs to parallelize one run.
# cpu_per_run=1,
)
sampler = AsyncSerialSampler(
EnvCls=AtariEnv,
TrajInfoCls=AtariTrajInfo,
env_kwargs=dict(game=game),
CollectorCls=DbCpuResetCollector,
batch_T=5,
batch_B=4,
max_decorrelation_steps=100,
eval_env_kwargs=dict(game=game),
eval_n_envs=1,
eval_max_steps=int(10e3),
eval_max_trajectories=2,
)
algo = DQN(replay_ratio=18, min_steps_learn=5e3, replay_size=int(1e5))
agent = AtariDqnAgent()
runner = AsyncRlEval(
algo=algo,
agent=agent,
sampler=sampler,
n_steps=2e6,
log_interval_steps=5e3,
affinity=affinity,
)
config = dict(game=game)
name = "async_dqn_" + game
log_dir = "async_dqn"
with logger_context(log_dir, run_ID, name, config):
runner.train()
def build_and_train(slot_affinity_code, log_dir, run_ID, config_key):
affinity = affinity_from_code(slot_affinity_code)
config = configs[config_key]
variant = load_variant(log_dir)
config = update_config(config, variant)
config["eval_env"]["game"] = config["env"]["game"]
sampler = AsyncSerialSampler(EnvCls=AtariEnv,
env_kwargs=config["env"],
CollectorCls=DbCpuResetCollector,
TrajInfoCls=AtariTrajInfo,
eval_env_kwargs=config["eval_env"],
**config["sampler"])
algo = DQN(optim_kwargs=config["optim"], **config["algo"])
agent = AtariDqnAgent(model_kwargs=config["model"], **config["agent"])
runner = AsyncRlEval(algo=algo,
agent=agent,
sampler=sampler,
affinity=affinity,
**config["runner"])
name = "async_serial_" + config["env"]["game"]
with logger_context(log_dir, run_ID, name, config):
runner.train()