def yield_runs():
# build solvers and train / test sets
if live:
domain = borg.get_domain(domain_name)
else:
domain = FakeDomain(borg.get_domain(domain_name))
paths = list(cargo.files_under(tasks_root, domain.extensions))
examples = int(round(len(paths) * 0.50))
logger.info("found %i tasks", len(paths))
if tests_root is not None:
tests_root_paths = list(cargo.files_under(tests_root, domain.extensions))
# build validation runs
for _ in xrange(runs):
split = uuid.uuid4()
shuffled = sorted(paths, key = lambda _ : numpy.random.rand())
train_paths = shuffled[:examples]
if tests_root is None:
test_paths = shuffled[examples:]
else:
test_paths = tests_root_paths
for name in borg.portfolios.named:
yield (run_validation, [name, domain, train_paths, test_paths, budget, split])
def __init__(self, tasks_roots, domain):
"""Initialize."""
# scan for CSV files
train_paths = []
for tasks_root in tasks_roots:
train_paths.extend(cargo.files_under(tasks_root, domain.extensions))
logger.info("using %i tasks for training", len(train_paths))
# fetch training data from each file
self._run_lists = {}
self._feature_vectors = {}
for path in train_paths:
# load run records
run_data = numpy.recfromcsv("{0}.runs.csv".format(path), usemask = True)
run_list = []
for (run_solver, run_budget, run_cost, run_succeeded, run_answer) in run_data.tolist():
record = RunRecord(run_solver, run_budget, run_cost, run_succeeded)
run_list.append(record)
self._run_lists[path] = run_list
# load feature data
feature_vector = numpy.recfromcsv("{0}.features.csv".format(path)).tolist()
self._feature_vectors[path] = feature_vector
def yield_runs():
bundle = borg.load_solvers(solvers_path)
paths = list(cargo.files_under(tasks_root, bundle.domain.extensions))
if not paths:
raise ValueError("no paths found under specified root")
if only_solver is None:
solver_names = bundle.solvers.keys()
else:
if only_solver not in bundle.solvers:
raise ArgumentError("no such solver")
solver_names = [only_solver]
for _ in xrange(runs):
for solver_name in solver_names:
for path in paths:
yield (run_solver_on, [solvers_path, solver_name, path, budget])
def yield_runs():
domain = borg.get_domain(domain_name)
paths = list(cargo.files_under(tasks_root, domain.extensions))
for path in paths:
yield (features_for_path, [domain, path])
