def load_data(args):
train_data_e2t, validation_data_e2t, test_data_e2t, \
train_data_ere, validation_data_ere, test_data_ere, \
train_data_trt, validation_data_trt, test_data_trt, \
entity2id, relation2id, type2id, headTailSelector, unique_entities_train_e2t, unique_types_train_e2t,\
unique_entities_train_ere, unique_types_train_trt= build_data(
args.data, is_unweigted=False, directed=True)
if args.pretrained_emb:
entity_embeddings, relation_embeddings = init_embeddings(
os.path.join(args.data, 'entity2vec.txt'),
os.path.join(args.data, 'relation2vec.txt'))
print("Initialised relations and entities from TransE")
else:
entity_embeddings = np.random.randn(len(entity2id),
args.entity_embedding_size)
relation_embeddings = np.random.randn(len(relation2id),
args.entity_embedding_size)
rdf_relation_embeddings = np.random.randn(1, args.type_embedding_size)
type_embeddings = np.random.randn(len(type2id),
args.type_embedding_size)
# entity2typeMat = np.random.randn(
# args.type_embedding_size, args.entity_embedding_size)
print("Initialised relations and entities randomly")
corpus_e2t = Corpus_e2t(args, train_data_e2t, validation_data_e2t,
test_data_e2t, entity2id, type2id,
args.batch_size_gat, args.valid_invalid_ratio_gat,
unique_entities_train_e2t, unique_types_train_e2t,
args.get_2hop)
corpus_ere = Corpus(args, train_data_ere, validation_data_ere,
test_data_ere, entity2id, relation2id,
headTailSelector, args.batch_size_gat,
args.valid_invalid_ratio_gat,
unique_entities_train_ere, args.get_2hop)
corpus_trt = Corpus(args, train_data_trt, validation_data_trt,
test_data_trt, type2id, relation2id, headTailSelector,
args.batch_size_gat, args.valid_invalid_ratio_gat,
unique_types_train_trt, args.get_2hop)
return corpus_e2t, corpus_ere, corpus_trt, torch.FloatTensor(
entity_embeddings), torch.FloatTensor(
relation_embeddings), torch.FloatTensor(
rdf_relation_embeddings), torch.FloatTensor(type_embeddings)
def load_data(args):
train_data, validation_data, test_data, entity2id, relation2id, headTailSelector, unique_entities_train = build_data(
args.data, is_unweigted=False, directed=True)
if args.pretrained_emb:
entity_embeddings, relation_embeddings = init_embeddings(
os.path.join(args.data, 'entity2vec.txt'),
os.path.join(args.data, 'relation2vec.txt'))
print("Initialised relations and entities from TransE")
else:
entity_embeddings = np.random.randn(len(entity2id),
args.embedding_size)
relation_embeddings = np.random.randn(len(relation2id),
args.embedding_size)
print("Initialised relations and entities randomly")
corpus = Corpus(args, train_data, validation_data, test_data, entity2id,
relation2id, headTailSelector, args.batch_size_gat,
args.valid_invalid_ratio_gat, unique_entities_train,
args.get_2hop)
return corpus, torch.FloatTensor(entity_embeddings), torch.FloatTensor(
relation_embeddings)
def __init__(self, args):
self.args = args
self.directory = args.data
# avoid temptation to refactor below, it is not worth it ;(
train_data, validation_data, test_data, entity2id, relation2id, headTailSelector, unique_entities_train = build_data(args.data, is_unweigted=False, directed=True)
self.corpus = Corpus(args, train_data, validation_data, test_data,
entity2id, relation2id, headTailSelector,
args.batch_size, args.valid_invalid_ratio,
unique_entities_train, args.get_2hop)
def load_data(args):
train_data, validation_data, test_data, entity2id, relation2id, headTailSelector, unique_entities_train = build_data(
args.data, is_unweigted=False, directed=True)
if args.pretrained_emb:
entity_embeddings, relation_embeddings = init_embeddings(os.path.join(args.data, 'entity2vec.txt'),
os.path.join(args.data, 'relation2vec.txt'))
print("Initialised relations and entities from TransE")
else:
entity_embeddings = np.random.randn(
len(entity2id), args.embedding_size)
relation_embeddings = np.random.randn(
len(relation2id), args.embedding_size)
print("Initialised relations and entities randomly")
'''
entity_embeddings = np.random.randn(
len(entity2id), args.embedding_size)
relation_embeddings = np.random.randn(
len(relation2id), args.embedding_size)
print("Initialised relations and entities randomly")
'''
corpus = Corpus(args, train_data, validation_data, test_data, entity2id, relation2id, headTailSelector,
args.batch_size_gat, args.valid_invalid_ratio_gat, unique_entities_train, args.get_2hop)
if (args.get_2hop):
file = args.data + "/2hop.pickle"
with open(file, 'wb') as handle:
pickle.dump(corpus.node_neighbors_2hop, handle,
protocol=pickle.HIGHEST_PROTOCOL)
if (args.use_2hop):
print("Opening node_neighbors pickle object")
file = args.data + "/2hop.pickle"
with open(file, 'rb') as handle:
node_neighbors_2hop = pickle.load(handle)
# return corpus, torch.FloatTensor(entity_embeddings), torch.FloatTensor(relation_embeddings)
return corpus, torch.cuda.FloatTensor(entity_embeddings), torch.cuda.FloatTensor(
relation_embeddings), node_neighbors_2hop
def load_data(args):
train_data, validation_data, test_data, entity2id, relation2id, args.id2entity, args.id2relation, headTailSelector, unique_entities_train, unique_relations_train = build_data(
args.data, is_unweigted=False, directed=False)
print('Training size', len(train_data), 'Val size', len(validation_data), 'Test size', len(test_data))
if args.pretrained_emb:
# no relation embedding for us now
entity_embeddings, relation_embeddings = init_embeddings(os.path.join(args.data, 'entity2vec.txt'),
None)
if entity_embeddings.shape[0] == 0:
entity_embeddings = np.random.randn(
len(entity2id), args.embedding_size)
if relation_embeddings.shape[0] == 0:
relation_embeddings = np.random.randn(
len(relation2id), args.embedding_size)
print("Initialised relations and entities from SSP")
else:
entity_embeddings = np.random.randn(
len(entity2id), args.embedding_size)
relation_embeddings = np.random.randn(
len(relation2id), args.embedding_size)
print("Initialised relations and entities randomly")
corpus = Corpus(args, train_data, validation_data, test_data, entity2id, relation2id, headTailSelector,
args.batch_size_gat, args.valid_invalid_ratio_gat, unique_entities_train, unique_relations_train, args.get_2hop)
return corpus, torch.FloatTensor(entity_embeddings), torch.FloatTensor(relation_embeddings)
