num=20,
base=2).astype('int') for start, end in _budgets
]
results = []
original_results = []
print(ss)
for s, budgets in zip(ss, _budgets):
s_results = []
for budget in budgets:
filename = 'data/HooVer_%s_nqueries_regret_budget%d_s%lf_exp%d.pklz' % (
model, budget, s, exp_id)
os.system(
'cd ../; python3 check.py --nRuns 1 --sigma 1e-5 --model %s --args %lf --budget %d --output %s --seed %d'
% (model, s, budget, filename, exp_id * 1024))
nimc = models.__dict__[model](s=s)
initial_states = loadpklz('../' + filename)['optimal_xs'][0]
original_results.append(
loadpklz('../' + filename)['optimal_values'][0])
result = nimc.get_prob(initial_states)
s_results.append(result)
results.append(s_results)
savepklz(
{
'results': results,
'ss': ss,
'budgets': _budgets,
'original_results': original_results
}, '../data/HooVer_%s_nqueries_regret_exp%d.pklz' % (model, exp_id))
# from IPython import embed; embed()
from utils.general_utils import loadpklz, savepklz
# -----------------------------------------------------------------------------
model = 'Slplatoon'
if __name__ == '__main__':
budget = 800000
outputs = []
bss = [10, 20, 50, 100, 400, 1600, 6400, 25600]
for exp_id in range(1, 10):
outputs.append([])
for bs in bss:
filename = 'data/HooVer_%s_budget%d_bs%d_exp%d.pklz' % (
model, budget, bs, exp_id)
outputs[-1].append(loadpklz('../' + filename))
depth = [[o['depths'][0] for o in exp] for exp in outputs]
mem = [[o['memory_usages'][0] for o in exp] for exp in outputs]
time = [[o['running_times'][0] for o in exp] for exp in outputs]
num_nodes = [[o['num_nodes'][0] for o in exp] for exp in outputs]
depth = np.array(depth).mean(axis=0)
mem = np.array(mem).mean(axis=0)
time = np.array(time).mean(axis=0)
num_nodes = np.array(num_nodes).mean(axis=0)
results_ho = []
for exp_id in range(1, 10):
results_ho.append(
loadpklz('../data/HooVer_%s_bs_exp%d.pklz' %
plt.rc('ytick', labelsize=15) # fontsize of the tick labels
plt.rc('legend', fontsize=20) # legend fontsize
plt.rc('figure', titlesize=BIGGER_SIZE) # fontsize of the figure title
plt.rc('axes', axisbelow=True)
left = 0.17 # the left side of the subplots of the figure
right = 0.98 # the right side of the subplots of the figure
bottom = 0.17 # the bottom of the subplots of the figure
top = 0.98 # the top of the subplots of the figure
model = 'ConceptualModel'
if __name__ == '__main__':
labels = ['s=0.1', 's=0.01', 's=0.001', 's=0.0003']
exps_ho = [
loadpklz('../data/HooVer_%s_nqueries_regret_exp%d.pklz' %
(model, exp_id)) for exp_id in range(1, 11)
]
ss = exps_ho[0]['ss']
budgets_ho = exps_ho[0]['budgets']
results_ho = [
np.stack([exp['results'][ids] for exp in exps_ho]).mean(axis=0)
for ids in range(len(ss))
]
exps_pl = [
loadpklz('../data/PlasmaLab_%s_nqueries_regret_exp%d.pklz' %
(model, exp_id)) for exp_id in range(1, 11)
]
ss = exps_pl[0]['ss']
budgets_pl = exps_pl[0]['budgets']
results_pl = [
import models
# -----------------------------------------------------------------------------
model = 'Mlplatoon'
exp_id = int(sys.argv[1])
if __name__ == '__main__':
budgets = np.logspace(np.log(0.85 * 1e5)/np.log(2), np.log(8e5)/np.log(2), num=6, base=2).astype('int')[:-1]
outputs = []
print('budgets: ' + str(budgets))
# run an experiment for each budget configuration
for budget in budgets:
filename = 'data/HooVer_%s_budget%d_exp%d.pklz'%(model, budget, exp_id)
os.system('cd ../; python3 check.py --nRuns 1 --sigma 1e-5 --model %s --budget %d --output %s --seed %d'%(model, budget, filename, exp_id*1024))
outputs.append(loadpklz('../'+filename))
results = []
optimal_xs = [o['optimal_xs'][0] for o in outputs]
original_results = [o['optimal_values'][0] for o in outputs]
num_queries = [o['budget'] for o in outputs]
# Monte-Carlo estimation of the hitting probability (using 250k simulations)
nimc = models.__dict__[model]()
for initial_states in optimal_xs:
initial_states = initial_states.tolist()
np.random.seed(1024)
result = evaluate_single_state(nimc, initial_states, nimc.k, mult=250000)
results.append(result)
print(result)
### enumerate all mu
def func(arg):
context = arg[1:]
arm = arg[0]
reward = []
np.random.seed(1024)
mult = 30000
for _ in range(mult):
reward.append(cmab.play_context(arm, context))
return np.mean(reward)
if os.path.exists('oracle.pklz'):
data = loadpklz('oracle.pklz')
args = data['args']
mu = data['mu']
else:
args = np.meshgrid(range(cmab.nArms), range(cmab.nZ), range(cmab.nU))
args = np.stack([arg.reshape(-1) for arg in args]).T
args = [args[i] for i in range(args.shape[0])]
with Pool(processes=25) as pool:
mu = list(tqdm.tqdm(pool.imap(func, args), total=len(args)))
savepklz({'mu': mu, 'args': args}, 'oracle.pklz')
# import ipdb; ipdb.set_trace()
# func(args[0])
args = np.stack(args)
left = 0.17 # the left side of the subplots of the figure
right = 0.98 # the right side of the subplots of the figure
bottom = 0.17 # the bottom of the subplots of the figure
top = 0.98 # the top of the subplots of the figure
# -----------------------------------------------------------------------------
model = 'Slplatoon'
if __name__ == '__main__':
budgets = np.logspace(np.log(0.85 * 1e5)/np.log(2), np.log(8e5)/np.log(2), num=6, base=2).astype('int')
outputs = []
num_nodes = []
# bss = [1, 10, 20, 50, 80, 100, 200, 400, 800]
bss = [10, 20, 50, 80, 100, 200, 400, 800]
exps = [loadpklz('../data/HooVer_%s_nqueries_regret_bs_exp%d.pklz'%(model, exp_id)) for exp_id in range(1, 11)]
data = np.zeros((len(bss), len(budgets), len(exps)))
budgets = budgets / 1e5
for budget in range(len(budgets)):
for bs in range(len(bss)):
for exp_id in range(1, 11):
data[bs, budget, exp_id-1] = exps[exp_id-1]['results'][budget*len(bss)+bs]
for ibs, bs in enumerate(bss):
plt.plot(budgets, 1-data.mean(axis=2)[ibs, :], 'o-', label='bs=%d'%bs)
plt.text(budgets[0], 1-data.mean(axis=2)[ibs, :][0], 'bs=%d'%bs)
plt.text(budgets[-1], 1-data.mean(axis=2)[ibs, :][-1], 'bs=%d'%bs)
plt.xlabel('#queries (x $10^5$)')
plt.ylabel('regret')
'$\mathsf{SLplatoon}(d=4, k=11)$', '$\mathsf{MLplatoon}(d=8, k=9)$',
'$\mathsf{DetectBrake}(d=4 , k=10)$', '$\mathsf{Merging}(d=4, k=10)$',
'$\mathsf{MLplatoon}(d=9, k=9)$'
]
models = [
models[int(sys.argv[1]) - 1],
]
for model in models:
results_pl = []
num_queries_pl = []
results_ho = []
num_queries_ho = []
for exp_id in range(1, 10):
results_pl.append(
loadpklz('../data/PlasmaLab_%s_exp%d.pklz' %
(model, exp_id))['results'])
num_queries_pl.append(
loadpklz('../data/PlasmaLab_%s_exp%d.pklz' %
(model, exp_id))['num_queries'])
results_ho.append(
loadpklz('../data/HooVer_%s_exp%d.pklz' %
(model, exp_id))['results'])
num_queries_ho.append(
loadpklz('../data/HooVer_%s_exp%d.pklz' %
(model, exp_id))['num_queries'])
results_pl = np.array(results_pl)
num_queries_pl = np.array(num_queries_pl) / 1e5
results_ho = np.array(results_ho)
num_queries_ho = np.array(num_queries_ho) / 1e5