has_entropy = (work_dir / '{}_0/policy_entropies.pkl'.format(dir_name)).exists()
print('has entropy:', has_entropy)
Js = list()
Rs = list()
Qs = list()
Es = list()
best_J = float("-inf")
best_stats = None
best_agent = None
skip_cnt = 0
for run_dir in run_dirs:
logger = BenchmarkLogger(log_dir=str(run_dir.parent), log_id=str(run_dir.name), use_timestamp=False)
print(run_dir, end=' ')
if not logger.exists_best_agent():
print("ABORT")
skip_cnt += 1
continue
else:
print("EXISTS")
Js.extend(logger.load_Js())
Rs.extend(logger.load_Rs())
Qs.extend(logger.load_Qs())
if has_entropy:
Es.extend(logger.load_policy_entropies())
stats = logger.load_stats()
if stats['best_J'] > best_J:
best_stats = stats
import os
from mushroom_rl_benchmark import BenchmarkLogger
from mushroom_rl_benchmark.experiment import exec_run
from mushroom_rl_benchmark.experiment.slurm import read_arguments_run
if __name__ == '__main__':
log_dir, run_args = read_arguments_run()
log_id = 'run_' + str(run_args['seed'])
agent_builder = BenchmarkLogger._load_pickle(
os.path.join(log_dir, 'agent_builder.pkl'))
env_builder = BenchmarkLogger._load_pickle(
os.path.join(log_dir, 'environment_builder.pkl'))
logger = BenchmarkLogger(log_dir=log_dir,
log_id=log_id,
use_timestamp=False)
logger.info('Starting experiment.')
result = exec_run(agent_builder, env_builder, **run_args)
logger.info('Saving result.')
cmp_E = agent_builder.compute_policy_entropy
logger.save_Js([result['Js']])
logger.save_Rs([result['Rs']])
import yaml
from pathlib import Path
from argparse import ArgumentParser
from mushroom_rl_benchmark import BenchmarkSuiteVisualizer, BenchmarkLogger
def get_args():
parser = ArgumentParser()
parser.add_argument(
"-d",
"--directory",
type=str,
required=True,
help='Benchmark directory where the plots generation is needed')
args = vars(parser.parse_args())
return args.values()
if __name__ == '__main__':
path, = get_args()
plots_file = Path('cfg') / 'plots.yaml'
logger = BenchmarkLogger.from_path(path)
with open(plots_file, 'r') as plots_file:
plot_params = yaml.safe_load(plots_file)
visualizer = BenchmarkSuiteVisualizer(logger, **plot_params)
visualizer.show_reports()
visualizer.save_reports()
"""
Script for benchmarking A2C Agent
"""
import time
from mushroom_rl_benchmark import BenchmarkExperiment, BenchmarkLogger
from mushroom_rl_benchmark.builders import EnvironmentBuilder, A2CBuilder
if __name__ == '__main__':
logger = BenchmarkLogger(log_dir='./logs', log_id='a2c_pendulum')
agent_builder = A2CBuilder.default(actor_lr=7e-4,
critic_lr=7e-4,
n_features=32)
env_name = 'Gym'
env_params = dict(env_id='Pendulum-v0', horizon=200, gamma=.99)
parallel = dict(max_concurrent_runs=10)
env_builder = EnvironmentBuilder(env_name, env_params)
logger.info('Environment is imported')
exp = BenchmarkExperiment(agent_builder, env_builder, logger)
logger.info('BenchmarkExperiment was built successfully')
start_time = time.time()
exp.run(exec_type='parallel',
n_runs=10,
n_epochs=100,
def run(res_dir, res_id):
"""
Function to aggregate the benchmark results from running in SLURM mode.
Args:
res_dir (str): path to the result directory;
res_id (str): log id of the result directory.
"""
work_dir = Path(res_dir, res_id)
# check if results are aggregated
dir_name = 'run'
run_dirs = list(work_dir.glob('{}_*'.format(dir_name)))
print(run_dirs)
has_entropy = (work_dir / '{}_0/entropy.pkl'.format(dir_name)).exists()
print('has entropy:', has_entropy)
J = list()
R = list()
V = list()
E = list()
best_J = float("-inf")
best_stats = None
best_agent = None
skip_cnt = 0
for run_dir in run_dirs:
logger = BenchmarkLogger(log_dir=str(run_dir.parent), log_id=str(run_dir.name), use_timestamp=False)
print(run_dir, end=' ')
if not logger.exists_best_agent():
print("ABORT")
skip_cnt += 1
continue
else:
print("EXISTS")
J.extend(logger.load_J())
R.extend(logger.load_R())
V.extend(logger.load_V())
if has_entropy:
E.extend(logger.load_entropy())
stats = logger.load_stats()
if stats['best_J'] > best_J:
best_stats = stats
best_agent = logger.load_best_agent()
if skip_cnt > 0: print('NUMBER OF FAILED RUNS:', '{}/{}'.format(skip_cnt, len(run_dirs)))
print('Name:', res_id)
print('Directory:', res_dir)
logger = BenchmarkLogger(log_dir=res_dir, log_id=res_id, use_timestamp=False)
logger.save_J(J)
logger.save_R(R)
logger.save_V(V)
if has_entropy:
logger.save_entropy(E)
logger.save_stats(best_stats)
logger.save_best_agent(best_agent)
visualizer = BenchmarkVisualizer(logger)
visualizer.save_report()
