def get_formulas(qrole, filename):
auto = True if "auto" in filename else False
data = utils.load_flattened_data(filename)
corr_mat = np.asarray(data[1:])
role2idx = {data[0][idx]: idx for idx in range(len(data[0]))}
idx2role = {idx: data[0][idx] for idx in range(len(data[0]))}
formulas = set()
qidx = role2idx[qrole]
corr_col = corr_mat[:,qidx]
if auto: # auto select formula for top 2 of matrix
bidx = np.argpartition(corr_col, -1)[-1:]
else: # select all manually set formulas in matrix (non-zero)
bidx = {}
for idx in range(corr_col.shape[0]):
formula_id = int(corr_col[idx])
if formula_id > 0:
if formula_id in bidx:
bidx[formula_id].append(idx)
else:
bidx[formula_id] = [idx]
for fidx in bidx.keys():
formula = []
for pidx in bidx[fidx]:
formula += [idx2role[pidx]]
formula += [qrole]
formula = tuple(sorted(formula))
formulas.add(formula)
return formulas
def get_formulas(filename):
auto = True if "auto" in filename else False
data = utils.load_flattened_data(filename)
corr_mat = np.asarray(data[1:])
role2idx = {data[0][idx]: idx for idx in range(len(data[0]))}
idx2role = {idx: data[0][idx] for idx in range(len(data[0]))}
formulas = set()
for qrole, qidx in role2idx.items():
corr_col = corr_mat[:, qidx]
if auto: # auto select formula for top 2 of matrix
bidx = np.argpartition(corr_col, -1)[-1:]
else: # select all manually set formulas in matrix (non-zero)
bidx = [
idx for idx in range(corr_col.shape[0]) if corr_col[idx] > 0.0
]
for idx in bidx:
formula = tuple(sorted([idx2role[idx]] + [qrole]))
formulas.add(formula)
return formulas
type=str,
help="(.txt)",
nargs="?",
default="./data/val_data_negative.txt")
parser.add_argument("--roles_file",
type=str,
help="(.txt)",
nargs="?",
default="./data/role_to_values.txt")
args = parser.parse_args()
# loads the MLN, DBs, and instances
with open(args.roles_file, "r") as f:
roles = eval(f.readlines()[0])
mln = MLN.load(args.input_mln)
dbs = Database.load(mln, args.positive_database)
p_examples = utils.load_flattened_data(args.positive_dataset)
n_examples = utils.load_flattened_data(args.negative_dataset)
test_times = []
# begins testing roles
for role in roles.keys():
start = time()
# creates testing DBs with labels
test_dbs = generate_test_dbs(role, dbs)
# gets MLN scores
scores = score_mln(mln, role, test_dbs)
# makes instance-score datastructure
instance_scores = scores2instance_scores(role, roles, p_examples,
n_examples, scores)
# gets metrics for the role
utils.compute_metric_scores(p_examples,
n_examples,
parser.add_argument("--formula_file",
type=str,
help="(.mx)",
nargs="?",
default="./data/formula_matrix_auto.mx")
parser.add_argument("--output_mln",
type=str,
help="(.mln)",
nargs="?",
default="./models/initial.mln")
args = parser.parse_args()
# loads the data for MLN
roles = utils.load_roles(args.roles_file)
instances = []
for dataset in args.input_datasets:
instances += utils.load_flattened_data(dataset)
role_constraints = utils.get_role_constraints(roles, instances)
formulas = get_formulas(args.formula_file)
domains = get_domains(roles, instances)
# generates the markov logic network
mln = MLN(logic="FirstOrderLogic", grammar="PRACGrammar")
for domain, values in domains.items(): # domains
for value in values:
if len(value) > 1:
const = ''.join(value)
elif len(value) > 0:
const = value[0]
else:
const = "None"
mln.update_domain({domain: [const]})
for role in roles.keys(): # predicates
instance.append(tuple((role, values)))
return mlninstances
if __name__ == "__main__":
parser = ArgumentParser(description="Role-Value Dataset 2 MLN Database")
parser.add_argument("--input_mln", type=str, help="(.mln)", nargs="?",
default="./models/class_initial.mln")
parser.add_argument("--input_datasets", type=str, help="(.txt)", nargs="*",
default=["./data/train_data.txt"])
parser.add_argument("--output_database", type=str, help="(.db)", nargs="?",
default="./data/train.db")
parser.add_argument("--roles_file", type=str, help="(.txt)", nargs="?",
default="./data/role_to_values.txt")
args = parser.parse_args()
# loads the initial MLN
mln = MLN.load(args.input_mln)
# loads data for DBs
atoms = []
roles = utils.load_roles(args.roles_file)
for input_dataset in args.input_datasets:
rv = utils.load_flattened_data(input_dataset)
rv = rvs2mlnrvs(mln, roles, rv)
# formats role-value to atoms
atoms += utils.format_instances_rv2atoms(rv)
# generates the DBs and saves
dbs = generate_databases(mln, atoms)
with open(args.output_database, "w") as f:
Database.write_dbs(dbs, f)
print("The database for the MLN is in " + args.output_database + ".")