#tmpl4_predictions = np.array([4]*len(mapped_data))
#tmpl5_predictions = np.array([5]*len(mapped_data))
#tmpl6_predictions = np.array([6]*len(mapped_data))
#if(args.rule_pred is not None):
# rule_predictions = utils.read_pkl(args.rule_pred)
# logging.info("Loaded rule predictions from %s" % (args.rule_pred))
# if(len(rule_predictions) != len(mapped_data)):
# logging.error("Unequal length of rule predictions and data")
# exit(-1)
entity_inverse_map = utils.get_inverse_dict(distmult_dump['entity_to_id'])
relation_inverse_map = utils.get_inverse_dict(
distmult_dump['relation_to_id'])
template_objs = template_builder.template_obj_builder(
data_root, args.model_weights, args.template_load_dir, None, "distmult", [1, 2, 3, 4, 5, 6], True)
explainer = Explainer(
data_root, template_objs[0].kb, template_objs[0].base_model, entity_inverse_map, relation_inverse_map)
#if(args.template_pred is not None):
template_exps = [english_exp_template(
mapped_data, var, template_objs, explainer) for var in template_predictions]
#else:
# template_exps = [
# explainer.NO_EXPLANATION for _ in range(len(mapped_data))]
#if(args.rule_pred is not None):
# rule_exps = english_exp_rules(mapped_data, rule_predictions, explainer)
#else:
# rule_exps = [explainer.NO_EXPLANATION for _ in range(len(mapped_data))]
logging.error("Unequal length of template predictions and data")
exit(-1)
if (args.rule_pred is not None):
rule_predictions = utils.read_pkl(args.rule_pred)
logging.info("Loaded rule predictions from %s" % (args.rule_pred))
if (len(rule_predictions) != len(mapped_data)):
logging.error("Unequal length of rule predictions and data")
exit(-1)
entity_inverse_map = utils.get_inverse_dict(distmult_dump['entity_to_id'])
relation_inverse_map = utils.get_inverse_dict(
distmult_dump['relation_to_id'])
template_objs = template_builder.template_obj_builder(
data_root, args.model_weights, args.template_load_dir, None,
"distmult", args.t_ids, True)
explainer = Explainer(data_root,
template_objs[0].kb,
template_objs[0].base_model,
entity_inverse_map,
relation_inverse_map,
list_of_different_template_files=['template1.txt'])
if (args.template_pred is not None):
template_exps = english_exp_template(mapped_data, template_predictions,
template_objs, explainer)
else:
template_exps = [
explainer.NO_EXPLANATION for _ in range(len(mapped_data))
utils._LOG_LEVEL_STRINGS))
args = parser.parse_args()
logging.basicConfig(format='%(levelname)s :: %(asctime)s - %(message)s',
level=args.log_level,
datefmt='%d/%m/%Y %I:%M:%S %p')
if (args.y_labels != '' and args.negative_count != 0):
logging.error(
'Cannot generate random samples with y labels. If using --y_labels use flag --negative_count 0 also'
)
exit(-1)
dataset_root = os.path.join(args.data_repo_root, args.dataset)
template_objs = template_builder.template_obj_builder(
dataset_root, args.model_weights, args.template_load_dir, None,
args.model_type, args.t_ids, args.oov_entity)
ktrain = template_objs[0].kb
k_preprocess = kb.KnowledgeBase(args.preprocess_file,
ktrain.entity_map,
ktrain.relation_map,
add_unknowns=not args.oov_entity)
y_labels = [1 for _ in range(k_preprocess.facts.shape[0])]
if (args.y_labels != ''):
#y_labels = np.loadtxt(args.y_labels)
y_labels, y_multilabels = utils.read_multilabel(args.y_labels)
if (y_labels.shape[0] != k_preprocess.facts.shape[0]):