def receiveResults(self):
resultsReceived = results_pb2.Results()
print('Waiting results...')
print(self.receiver)
message = self.receiver.recv()
print('Got results')
resultsReceived.ParseFromString(message)
self.resultQueue.put(resultsReceived)
def load_results(result_directory): # noqa
data = {}
results = results_pb2.Results()
for filename in os.listdir(result_directory):
with open(os.path.join(result_directory, filename), 'rb') as f:
try:
results.ParseFromString(f.read())
except Exception as e:
# Checking if is cluster stderr file.
if filename.endswith('.err'):
continue
raise e
for method in results.methods:
mse_key = '%s_mse' % method.method_name
if mse_key not in data:
data[mse_key] = []
data[mse_key].append(np.array(method.mse))
return data
instances = set()
lines = sys.stdin.readlines()
idx = 0
# read the results
while idx < len(lines):
words = lines[idx].strip().split()
n = float(words[0])
rho = float(words[1])
inst = words[2]
nAnts.add(n)
rhos.add(rho)
instances.add(inst)
rs = []
for rf in words[3:]:
rb = results_pb2.Results()
with open(rf, "rb") as f:
rb.ParseFromString(f.read())
rs.append(rb)
results[(inst, n, rho)] = rs
idx = idx + 1
outfile = open(sys.argv[1] + '/aco-heatmap', 'w')
outfile.write('nAnts rho')
for inst in range(len(instances)):
outfile.write(' ' + 'inst' + str(inst))
outfile.write('\n')
for n in nAnts:
for rho in rhos:
outfile.write(str(n) + ' ' + str(rho))
def timeToBaseQ(s):
fracs=[]
for (i,(b,r)) in results.iteritems():
baseR = b.point[-1]
fracs.append( float(timeTo(baseR.qual,r[s])) / float(baseR.time) )
return sum(fracs) / float(len(fracs))
idx=0
while idx < len(lines):
# read an instance
instanceName = lines[idx].strip()
print "reading " + instanceName
idx=idx+1
instanceBaseLineFile = lines[idx].strip()
idx=idx+1
baseLineR = results_pb2.Results()
with open(instanceBaseLineFile, "rb") as f:
baseLineR.ParseFromString(f.read())
if len(baseLineR.point) > 0:
results[instanceName] = (baseLineR,{})
# read all alternative methods results
while idx < len(lines) and lines[idx].strip() != '':
solverName = lines[idx].strip()
solvers.add(solverName)
idx=idx+1
# TODO: Take an average if needed
solverResultsFile = lines[idx].strip()
idx=idx+1
solverResults = results_pb2.Results()
def main(): # noqa
if FLAGS.seed is not None:
np.random.seed(FLAGS.seed)
n_actions = 2
horizon = 5
n_methods = 2 * (horizon + 1)
mdp = mdps.SinglePathMDP(n_actions, horizon, stochastic=True)
pie = mdp.get_policy(0)
pib = mdp.get_policy(1)
print(pie)
print(pib)
paths = []
results = results_pb2.Results()
results.experiment_name = 'RIS_experiment'
t_val = mdps.evaluate(mdp, pie)
print('True value: %f' % t_val)
IS = estimators.ISEstimate(pie, pib)
WIS = estimators.ISEstimate(pie, pib, weighted=True)
REG = estimators.REGEstimate(pie, mdp)
methods = []
labels = []
for i in range(horizon):
n = i + 1
methods.append(estimators.RISEstimate(pie, n_actions, horizon, n=n))
methods.append(
estimators.RISEstimate(pie, n_actions, horizon, n=n,
weighted=True))
labels.append('RIS(%d)' % n)
labels.append('Weighted RIS(%d)' % n)
n_evals = int(FLAGS.num_iters / FLAGS.eval_freq)
is_mses = np.zeros(n_evals)
is_estimates = np.zeros(n_evals)
is_variances = np.zeros(n_evals)
wis_mses = np.zeros(n_evals)
wis_estimates = np.zeros(n_evals)
reg_mses = np.zeros(n_evals)
reg_estimates = np.zeros(n_evals)
mses = np.zeros((n_methods, n_evals))
variances = np.zeros((n_methods, n_evals))
estimates = np.zeros((n_methods, n_evals))
lens = []
idx = 0
for itr in range(FLAGS.num_iters):
path, G = mdps.sample(mdp, pib)
paths.append(path)
if itr % FLAGS.eval_freq == 0 and itr > 0:
# idx = int(itr / FLAGS.eval_freq)
eval_estimators(paths, idx, t_val, IS, is_estimates, is_mses,
is_variances, WIS, wis_estimates, wis_mses, REG,
reg_estimates, reg_mses, methods, estimates, mses,
variances, labels)
idx += 1
eval_estimators(paths, idx, t_val, IS, is_estimates, is_mses, is_variances,
WIS, wis_estimates, wis_mses, REG, reg_estimates, reg_mses,
methods, estimates, mses, variances, labels)
# Normal IS methods
method = results.methods.add()
method.method_name = 'IS'
method.estimates.extend(is_estimates)
method.mse.extend(is_mses)
method.variances.extend(is_variances)
method = results.methods.add()
method.method_name = 'WIS'
method.estimates.extend(wis_estimates)
method.mse.extend(wis_mses)
method = results.methods.add()
method.method_name = 'REG'
method.mse.extend(reg_mses)
method.estimates.extend(reg_estimates)
# Add RIS methods
for i in range(2 * horizon):
method = results.methods.add()
method.method_name = labels[i]
method.estimates.extend(estimates[i])
method.mse.extend(mses[i])
method.variances.extend(variances[i])
if FLAGS.result_file:
with open(FLAGS.result_file, 'wb') as w:
w.write(results.SerializeToString())
# Plotting code
if FLAGS.plot:
lens = np.arange(n_evals) * FLAGS.eval_freq
plt.plot(lens, is_mses, label='IS')
plt.plot(lens, wis_mses, label='WIS')
plt.plot(lens, reg_mses, label='REG')
for i in range(2 * horizon):
plt.plot(lens, mses[i], label=labels[i])
plt.legend()
plt.show()