def main():
if args.cuda and torch.cuda.is_available():
device = torch.device("cuda")
else:
device = torch.device("cpu")
print("Using device", type(device))
algorithmopts = yaml.safe_load(open(args.algorithm_specs, "r"))
testcase = load_example(algorithmopts["example"])
# start_time = time.time() #######
model, likelihood = init_gp(testcase, algorithmopts,
algorithmopts["ninit"], device)
# time1 = time.time() ####
# print("--- init_gp %s seconds ---" % (time1 - start_time)) ###
estimator = ExcursionSetEstimator(testcase, algorithmopts, model,
likelihood, device)
timestampStr = datetime.datetime.now().strftime("%d-%b-%Y_%H:%M:%S") + "/"
os.mkdir(args.outputfolder + timestampStr)
while estimator.this_iteration < algorithmopts["nupdates"]:
time2 = time.time()
estimator.step(testcase, algorithmopts, model, likelihood)
time3 = time.time() ####
print("--- step %s seconds ---" % (time3 - time2)) ###
model = estimator.update_posterior(testcase, algorithmopts, model,
likelihood)
time4 = time.time() ####
print("--- posterior %s seconds ---" % (time4 - time3)) ###
estimator.plot_status(testcase, algorithmopts, model,
estimator.acq_values,
args.outputfolder + timestampStr)
estimator.get_diagnostics(testcase, model, likelihood)
time5 = time.time() ####
print("--- get_diagnostics %s seconds ---" % (time5 - time4)) ###
estimator.print_results(args.outputfolder + timestampStr, testcase,
algorithmopts)
def test_init_gp():
device = torch.device("cpu")
ninit = 1
algorithmopts = yaml.safe_load(open("testing/algorithm_specs.yaml", "r"))
# three toy examples
for example in ["1Dtoyanalysis", "2Dtoyanalysis", "3Dtoyanalysis"]:
print(example)
testcase = load_example(example)
gp, likelihood = init_gp(testcase, algorithmopts, ninit, device)
assert type(gp.train_inputs[0]) != type(None)
assert type(gp.train_targets) != type(None)
def test_init_excursion():
device = torch.device("cpu")
ninit = 1
algorithmopts = yaml.safe_load(open("testing/algorithm_specs.yaml", "r"))
# three toy examples
for example in ["1Dtoyanalysis", "2Dtoyanalysis", "3Dtoyanalysis"]:
testcase = load_example(example)
gp, likelihood = init_gp(testcase, algorithmopts, ninit, device)
estimator = ExcursionSetEstimator(testcase, algorithmopts, gp,
likelihood, device)
assert type(estimator) != type(None)
assert type(estimator._X_grid) != type(None)
assert type(estimator._n_dims) != type(None)
assert type(estimator._acq_type) != type(None)
def test_full_simple():
tol = 1e-6
device = torch.device("cpu")
ninit = 1
algorithmopts = yaml.safe_load(
open("testing/algorithm_specs_full_test.yaml", "r"))
# three toy examples
for example in ["1D_test"]:
testcase = load_example(example)
model, likelihood = init_gp(testcase, algorithmopts, ninit, device)
estimator = ExcursionSetEstimator(testcase, algorithmopts, model,
likelihood, device)
while estimator.this_iteration < algorithmopts["nupdates"]:
estimator.step(testcase, algorithmopts, model, likelihood)
model = estimator.update_posterior(testcase, algorithmopts, model,
likelihood)
assert type(torch.abs(model.train_targets) <= tol) != type(None)
def test_step_excursion():
device = torch.device("cpu")
ninit = 1
algorithmopts = yaml.safe_load(open("testing/algorithm_specs.yaml", "r"))
# three toy examples
for example in ["1Dtoyanalysis", "2Dtoyanalysis", "3Dtoyanalysis"]:
testcase = load_example(example)
gp, likelihood = init_gp(testcase, algorithmopts, ninit, device)
estimator = ExcursionSetEstimator(testcase, algorithmopts, gp,
likelihood, device)
# one iteration only to test
estimator.step(testcase, algorithmopts, gp, likelihood)
assert type(estimator) != type(None)
assert type(estimator.x_new) != type(None)
assert type(estimator.y_new) != type(None)
assert estimator.walltime_step != 0.0
assert estimator.this_iteration == 1
plt.savefig('plots/test_fast2D.png')
#load algorithm options
file = open('excursion/testcases/algorithms/algorithm_specs.yaml', "r")
algorithm_opts = yaml.safe_load(file)
print('algorithm_opts =')
print(json.dumps(algorithm_opts, indent=4))
## Step 2: Init the Gaussian Proces
#We initialize the gaussian process and likelihood according to `algorithm_opts['init_type']`
if (torch.cuda.is_available()):
device = torch.device('cuda')
else:
device = torch.device('cpu')
model, likelihood = init_gp(testcase_details, \
algorithm_opts, \
algorithm_opts["ninit"], \
device)
# X_init and y_init are random points so that we can fit the GP posterior wrt these points
# fit hyperparameters
model.train()
likelihood.train()
excursion.fit_hyperparams(model, likelihood)
# plot the fitting
plots.plot_GP_init(plt,
model,
testcase_details,
device=device,
dtype=torch.float64)
## Step 3: Init the ExcursionEstimator
estimator = ExcursionSetEstimator(testcase_details, \