def black_box_function(bandwidth_lower_bound: float,
bandwidth_upper_bound: float, delay: float, queue: float,
loss: float, T_s: float, delay_noise: float, heuristic,
model_path: str, save_dir: str = "") -> float:
global black_box_function_calling_times
save_dir = os.path.join(save_dir, 'config_{}'.format(
black_box_function_calling_times % 15))
black_box_function_calling_times += 1
heuristic_rewards = []
rl_method_rewards = []
if loss < -4:
loss = 0
else:
loss = 10**loss
traces = [generate_trace(
duration_range=(30, 30),
bandwidth_lower_bound_range=(
10**bandwidth_lower_bound, 10**bandwidth_lower_bound),
bandwidth_upper_bound_range=(
10**bandwidth_upper_bound, 10**bandwidth_upper_bound),
delay_range=(delay, delay),
loss_rate_range=(loss, loss),
queue_size_range=(queue, queue),
T_s_range=(T_s, T_s),
delay_noise_range=(delay_noise, delay_noise)) for _ in range(10)]
# print("trace generation used {}s".format(time.time() - t_start))
save_dirs = [os.path.join(save_dir, 'trace_{}'.format(i)) for i in range(10)]
for trace, save_dir in zip(traces, save_dirs):
os.makedirs(save_dir, exist_ok=True)
trace.dump(os.path.join(save_dir, 'trace.json'))
if not heuristic:
for trace in traces:
heuristic_rewards.append(trace.optimal_reward)
else:
t_start = time.time()
# save_dirs = [""] * len(traces)
hret = heuristic.test_on_traces(traces, save_dirs, True, 8)
for heuristic_mi_level_reward, heuristic_pkt_level_reward in hret:
# heuristic_rewards.append(heuristic_mi_level_reward)
heuristic_rewards.append(heuristic_pkt_level_reward)
print("heuristic used {}s".format(time.time() - t_start))
t_start = time.time()
rl_ret = test_on_traces(model_path, traces, save_dirs, 8, 20,
record_pkt_log=False, plot_flag=True)
for rl_mi_level_reward, rl_pkt_level_reward in rl_ret:
# rl_method_rewards.append(rl_mi_level_reward)
rl_method_rewards.append(rl_pkt_level_reward)
print("rl used {}s".format(time.time() - t_start))
gap = float(np.mean(np.array(heuristic_rewards)) - np.mean(np.array(rl_method_rewards)))
return gap
def main():
args = parse_args()
set_seed(args.seed)
if args.heuristic == 'cubic':
cc = Cubic()
elif args.heuristic == 'bbr_old':
cc = BBR_old()
elif args.heuristic == 'bbr':
cc = BBR()
else:
raise NotImplementedError
# if 'large' in args.config_file:
# config = read_json_file(args.config_file)
# config[0]['bandwidth_lower_bound'] = (1, 1)
# config[0]['bandwidth_upper_bound'] = (1, 100)
# traces = generate_traces_from_config(config, 50, 30)
# else:
if not args.config_file:
dataset = PantheonDataset('../../data', 'all')
traces = dataset.get_traces(0, 50)
save_dirs = [os.path.join(
args.save_dir, link_conn_type, link_name, trace_name)
for link_conn_type, (link_name, trace_name) in
zip(dataset.link_conn_types, dataset.trace_names)]
else:
traces = generate_traces(args.config_file, 50, 30)
save_dirs = [os.path.join(args.save_dir, "trace_{:02d}".format(i))
for i in range(len(traces))]
cc_save_dirs = [os.path.join(save_dir, cc.cc_name)
for save_dir in save_dirs]
cc_res = cc.test_on_traces(traces, cc_save_dirs, plot_flag=False, n_proc=16)
aurora_save_dirs = [os.path.join(save_dir, 'aurora')
for save_dir in save_dirs]
aurora_res = test_on_traces(args.model_path, traces, aurora_save_dirs,
nproc=16, seed=42, record_pkt_log=False,
plot_flag=False)
print(cc.cc_name, np.mean([res[1] for res in cc_res]))
print('aurora', np.mean([res[1] for res in aurora_res]))
for i, (trace, save_dir) in enumerate(zip(traces, save_dirs)):
trace.dump(os.path.join(save_dir, 'trace_{:02d}.json'.format(i)))
def main():
args = parse_args()
set_seed(args.seed)
# tokens = os.path.basename(os.path.dirname(os.path.dirname(args.save_dir))).split('_')
# config0_dim0_idx = int(tokens[1])
# config0_dim1_idx = int(tokens[2])
# config1_dim0_idx = int(tokens[4])
# config1_dim1_idx = int(tokens[5])
dim0, dim1 = args.dims
config = read_json_file(args.config_file)[0]
assert dim0 in config and dim1 in config
# dim0_vals = np.linspace(config[dim0][0], config[dim0][1], 10)
# dim1_vals = np.linspace(config[dim1][0], config[dim1][1], 10)
dim0_vals = get_dim_vals(dim0)
dim1_vals = get_dim_vals(dim1)
print(dim0_vals)
print(dim1_vals)
traces = []
save_dirs = []
with open('heatmap_trace_cnt_ratio.npy', 'rb') as f:
cnt_ratio = np.load(f)
for dim0_idx, dim0_val in enumerate(dim0_vals):
for dim1_idx, dim1_val in enumerate(dim1_vals):
dim_vals = copy.copy(DEFAULT_VALUES)
dim_vals[dim0] = dim0_val
dim_vals[dim1] = dim1_val
# print(i, dim0_val, dim1_val, dim_vals)
cnt = 10
# if cnt_ratio[dim0_idx, dim1_idx] > 1:
# cnt *= int(cnt_ratio[dim0_idx, dim1_idx])
# print(cnt)
for trace_idx in range(cnt):
trace = generate_trace(
duration_range=(dim_vals['duration'],
dim_vals['duration']),
bandwidth_lower_bound_range=(
dim_vals['bandwidth_lower_bound'],
dim_vals['bandwidth_lower_bound']),
bandwidth_upper_bound_range=(
dim_vals['bandwidth_upper_bound'],
dim_vals['bandwidth_upper_bound']),
delay_range=(dim_vals['delay'], dim_vals['delay']),
loss_rate_range=(dim_vals['loss'], dim_vals['loss']),
queue_size_range=(dim_vals['queue'], dim_vals['queue']),
T_s_range=(dim_vals['T_s'], dim_vals['T_s']),
delay_noise_range=(dim_vals['delay_noise'],
dim_vals['delay_noise']))
traces.append(trace)
save_dir = os.path.join(
args.save_dir, 'pair_{}_{}'.format(dim0_idx, dim1_idx),
'trace_{}'.format(trace_idx))
save_dirs.append(save_dir)
os.makedirs(save_dir, exist_ok=True)
trace.dump(
os.path.join(save_dir, 'trace_{}.json'.format(trace_idx)))
if args.cc == 'genet_bbr' or args.cc == 'genet_cubic' or args.cc == 'genet_bbr_old':
genet_seed = ''
for s in args.models_path.split('/'):
if 'seed' in s:
genet_seed = s
for bo in range(0, 30, 3):
# for bo_dir in natural_sort(glob.glob(os.path.join(args.models_path, "bo_*/"))):
bo_dir = os.path.join(args.models_path, "bo_{}".format(bo))
step = 64800
model_path = os.path.join(bo_dir,
'model_step_{}.ckpt'.format(step))
if not os.path.exists(model_path + '.meta'):
print(model_path, 'does not exist')
continue
print(model_path)
genet_save_dirs = [
os.path.join(save_dir, args.cc, genet_seed, "bo_{}".format(bo),
"step_{}".format(step)) for save_dir in save_dirs
]
t_start = time.time()
test_on_traces(model_path, traces, genet_save_dirs, args.nproc, 42,
False, False)
print('bo {}: {:.3f}'.format(bo, time.time() - t_start))
elif args.cc == 'pretrained':
pretrained_save_dirs = [
os.path.join(save_dir, args.cc) for save_dir in save_dirs
]
t_start = time.time()
test_on_traces(args.models_path, traces, pretrained_save_dirs,
args.nproc, 42, False, False)
print('pretrained: {:.3f}'.format(time.time() - t_start))
elif args.cc == 'overfit_config':
overfit_config_save_dirs = [
os.path.join(save_dir, args.cc) for save_dir in save_dirs
]
t_start = time.time()
test_on_traces(args.models_path, traces, overfit_config_save_dirs,
args.nproc, 42, False, False)
print('overfit_config: {:.3f}'.format(time.time() - t_start))
else:
if args.cc == 'bbr':
cc = BBR(False)
elif args.cc == 'cubic':
cc = Cubic(False)
elif args.cc == 'bbr_old':
cc = BBR_old(False)
else:
raise NotImplementedError
heuristic_save_dirs = [
os.path.join(save_dir, cc.cc_name) for save_dir in save_dirs
]
t_start = time.time()
cc.test_on_traces(traces, heuristic_save_dirs, False, args.nproc)
print('{}: {:.3f}'.format(args.cc, time.time() - t_start))
def main():
args = parse_args()
# dataset = SyntheticDataset(25, '../../config/train/udr_7_dims_0826/udr_large.json', seed=42)
dataset = SyntheticDataset.load_from_dir(
'/datamirror/zxxia/PCC-RL/results_1006/synthetic_dataset')
traces = dataset.traces
# cnt = 0
# for idx, trace in enumerate(traces):
# if trace.min_bw > 0.5:
# cnt+=1
# print(idx)
# print(cnt)
# return
save_dirs = [
os.path.join(args.save_dir, 'trace_{:05d}'.format(i))
for i in range(len(dataset))
]
if args.cc == 'bbr':
cc = BBR(False)
cc.test_on_traces(
traces,
[os.path.join(save_dir, cc.cc_name)
for save_dir in save_dirs], True, args.nproc)
elif args.cc == 'bbr_old':
cc = BBR_old(False)
cc.test_on_traces(
traces,
[os.path.join(save_dir, cc.cc_name)
for save_dir in save_dirs], True, args.nproc)
elif args.cc == 'cubic':
cc = Cubic(False)
cc.test_on_traces(
traces,
[os.path.join(save_dir, cc.cc_name)
for save_dir in save_dirs], True, args.nproc)
elif args.cc == 'real':
model_path = args.models_path
real_save_dirs = [
os.path.join(save_dir, args.cc, "seed_{}".format(args.seed))
for save_dir in save_dirs
]
test_on_traces(model_path,
traces,
real_save_dirs,
args.nproc,
42,
False,
plot_flag=True)
elif args.cc == 'pretrained':
udr_seed = ''
for s in args.models_path.split('/'):
if 'seed' in s:
udr_seed = s
step = 7200
while step <= 151200:
if not os.path.exists(
os.path.join(args.models_path,
'model_step_{}.ckpt.meta'.format(step))):
break
udr_save_dirs = [
os.path.join(save_dir, args.cc, udr_seed,
"step_{}".format(step)) for save_dir in save_dirs
]
model_path = os.path.join(args.models_path,
'model_step_{}.ckpt'.format(step))
test_on_traces(model_path, traces, udr_save_dirs, args.nproc, 42,
False, True)
step += 28800
elif args.cc == 'udr1' or args.cc == 'udr2' or args.cc == 'udr3' or \
args.cc == 'cl1' or args.cc == 'cl1_new' or args.cc == 'cl2' or \
args.cc == 'cl2_new' or args.cc == 'real_cellular' or \
'udr' in args.cc or 'cl' in args.cc:
# TODO: bug here when there is no validation log
# original implementation
# val_log = pd.read_csv(os.path.join(
# args.models_path, 'validation_log.csv'), sep='\t')
# for idx in np.linspace(0, len(val_log['num_timesteps']) / 2 - 1, 10):
# step = int(val_log['num_timesteps'].iloc[int(idx)])
# udr_seed = ''
# for s in args.models_path.split('/'):
# if 'seed' in s:
# udr_seed = s
# udr_save_dirs = [os.path.join(
# save_dir, args.cc, udr_seed, "step_{}".format(step)) for save_dir in save_dirs]
# model_path = os.path.join(
# args.models_path, 'model_step_{}.ckpt'.format(step))
# test_on_traces(model_path, traces, udr_save_dirs,
# args.nproc, 42, False, False)
# new implementation to be fair with genet
udr_seed = ''
for s in args.models_path.split('/'):
if 'seed' in s:
udr_seed = s
step = 0
while step <= 720000:
if not os.path.exists(
os.path.join(args.models_path,
'model_step_{}.ckpt.meta'.format(step))):
break
udr_save_dirs = [
os.path.join(save_dir, args.cc, udr_seed,
"step_{}".format(step)) for save_dir in save_dirs
]
model_path = os.path.join(args.models_path,
'model_step_{}.ckpt'.format(step))
test_on_traces(model_path, traces, udr_save_dirs, args.nproc, 42,
False, True)
# step += (7200) * 2
step += 72000
print(step)
# elif args.cc == 'genet_bbr' or args.cc == 'genet_cubic' or 'genet_bbr_old':
elif 'genet' in args.cc: #== 'genet_bbr' or args.cc == 'genet_cubic' or 'genet_bbr_old': genet_seed = ''
for s in args.models_path.split('/'):
if 'seed' in s:
genet_seed = s
# for bo in range(0, 15, 3):
for bo in range(0, 10):
bo_dir = os.path.join(args.models_path, "bo_{}".format(bo))
for step in range(64800, 72000, 14400):
# step = 64800
model_path = os.path.join(bo_dir,
'model_step_{}.ckpt'.format(step))
if not os.path.exists(model_path + '.meta'):
print("skip " + model_path + '.meta')
continue
genet_save_dirs = [
os.path.join(save_dir, args.cc, genet_seed,
"bo_{}".format(bo), "step_{}".format(step))
for save_dir in save_dirs
]
# print(genet_save_dirs)
test_on_traces(model_path, traces, genet_save_dirs, args.nproc,
42, False, True)
else:
raise ValueError
def main():
args = parse_args()
dataset = SyntheticDataset(
1000, '../../config/train/udr_7_dims_0826/udr_large.json')
save_dirs = [
os.path.join(args.save_dir, 'trace_{:05d}'.format(i))
for i in range(len(dataset))
]
for seed in range(60, 70, 10):
bo = 9
step = 64800
# cc = BBR_old(False)
# cc.test_on_traces(dataset.traces, [os.path.join(save_dir, cc.cc_name)
# for save_dir in save_dirs], True, args.nproc)
# cc = BBR(False)
# cc.test_on_traces(dataset.traces, [os.path.join(save_dir, cc.cc_name)
# for save_dir in save_dirs], True, args.nproc)
# cc = Cubic(False)
# cc.test_on_traces(dataset.traces, [os.path.join(save_dir, cc.cc_name)
# for save_dir in save_dirs], True, args.nproc)
# break
# model_path = "/datamirror/zxxia/PCC-RL/results_1006/genet_better_bw/genet_bbr_old/seed_{}/bo_{}/model_step_{}.ckpt".format(seed, bo, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'genet_bbr_old_better_bw', "seed_{}".format(seed), "bo_{}".format(bo),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, False)
# model_path = "/datamirror/zxxia/PCC-RL/results_1006/genet_random_real/0.05/seed_{}/bo_{}/model_step_{}.ckpt".format(seed, bo, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'genet_bbr_old_5percent_real', "seed_{}".format(seed), "bo_{}".format(bo),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, False)
# model_path = "/tank/zxxia/PCC-RL/results_0928/genet_no_reward_scale/genet_bbr_old/seed_{}/bo_{}/model_step_{}.ckpt".format(seed, bo, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'genet_bbr_old_cpu', "seed_{}".format(seed), "bo_{}".format(bo),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, True)
# step = 705600
# model_path = "/datamirror/zxxia/PCC-RL/results_1006/train_with_difficulty_metrics/metric1_new/seed_{}/model_step_{}.ckpt".format(seed, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'cl1', "seed_{}".format(seed),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, False)
# model_path = "/datamirror/zxxia/PCC-RL/results_1006/train_with_difficulty_metrics/metric2_cubic/seed_{}/model_step_{}.ckpt".format(seed, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'cl2', "seed_{}".format(seed),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, False)
step = 720000
# model_path = "/tank/zxxia/PCC-RL/results_0905/udr/udr_small/seed_{}/model_step_{}.ckpt".format(seed, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'udr1', "seed_{}".format(seed),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, False)
# model_path = "/tank/zxxia/PCC-RL/results_0928/udr1/seed_{}/model_step_{}.ckpt".format(seed, step)
# genet_save_dirs = [os.path.join(
# save_dir, 'udr1_1023', "seed_{}".format(seed),
# "step_{}".format(step)) for save_dir in save_dirs]
# test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc, 42, False, False)
model_path = "/tank/zxxia/PCC-RL/results_0928/udr2/seed_{}/model_step_{}.ckpt".format(
seed, step)
genet_save_dirs = [
os.path.join(save_dir, 'udr2_1023', "seed_{}".format(seed),
"step_{}".format(step)) for save_dir in save_dirs
]
test_on_traces(model_path, dataset.traces, genet_save_dirs, args.nproc,
42, False, False)
'trace_{:05d}'.format(i)) for i in range(len(traces))
]
before_save_dirs = [
os.path.join(save_dir, 'before') for save_dir in save_dirs
]
after_save_dirs = [
os.path.join(save_dir, 'after_best_pkt_level_reward')
for save_dir in save_dirs
]
bbr_old_save_dirs = [
os.path.join(save_dir, 'bbr_old') for save_dir in save_dirs
]
if os.path.exists(os.path.join(after_save_dirs[0], 'aurora_summary.csv')):
continue
test_on_traces(MODEL_PATH_BEFORE, traces, before_save_dirs, 16, seed,
False, True)
test_on_traces(model_path_after, traces, after_save_dirs, 16, seed, False,
True)
after_save_dirs = [
os.path.join(save_dir, 'after') for save_dir in save_dirs
]
model_path_after = os.path.join(ROOT, 'seed_{}'.format(seed),
'config_{:02d}'.format(config_id),
'model_step_{}.ckpt'.format(64800))
test_on_traces(model_path_after, traces, after_save_dirs, 16, seed, False,
True)
baseline = BBR_old(False)
baseline.test_on_traces(traces,
bbr_old_save_dirs,
plot_flag=True,