def test(dataset, args, all_hash, modelD, tflogger, subsample=1):
l_ranks, r_ranks = [], []
test_loader = DataLoader(dataset, num_workers=1, collate_fn=collate_fn)
cst_inds = np.arange(args.num_ent, dtype=np.int64)[:, None]
if args.show_tqdm:
data_itr = tqdm(enumerate(test_loader))
else:
data_itr = enumerate(test_loader)
for idx, triplet in data_itr:
if idx % subsample != 0:
continue
lhs, rel, rhs = triplet.view(-1)
l_batch = np.concatenate(
[cst_inds,
np.array([[rel, rhs]]).repeat(args.num_ent, axis=0)],
axis=1)
r_batch = np.concatenate(
[np.array([[lhs, rel]]).repeat(args.num_ent, axis=0), cst_inds],
axis=1)
l_fns = np.array([int(x.tobytes() in all_hash) for x in l_batch],
dtype=np.float32)
r_fns = np.array([int(x.tobytes() in all_hash) for x in r_batch],
dtype=np.float32)
l_batch = ltensor(l_batch).contiguous()
r_batch = ltensor(r_batch).contiguous()
if args.use_cuda:
l_batch = l_batch.cuda()
r_batch = r_batch.cuda()
l_batch = Variable(l_batch)
r_batch = Variable(r_batch)
d_ins = torch.cat([l_batch, r_batch], dim=0)
d_outs = modelD(d_ins)
l_enrgs = d_outs[:len(l_batch)]
r_enrgs = d_outs[len(l_batch):]
l_rank = compute_rank(v2np(l_enrgs), lhs, mask_observed=l_fns)
r_rank = compute_rank(v2np(r_enrgs), rhs, mask_observed=r_fns)
l_ranks.append(l_rank)
r_ranks.append(r_rank)
l_ranks = np.array(l_ranks)
r_ranks = np.array(r_ranks)
return l_ranks, r_ranks
def test(dataset):
l_ranks, r_ranks = [], []
test_loader = DataLoader(dataset, num_workers=1, collate_fn=collate_fn)
cst_inds = np.arange(args.num_ent, dtype=np.int64)[:, None]
for idx, triplet in tqdm(enumerate(test_loader)):
lhs, rel, rhs = triplet.view(-1)
l_batch = np.concatenate([
cst_inds,
np.array([[rel, rhs]]).repeat(args.num_ent, axis=0)
],
axis=1)
r_batch = np.concatenate([
np.array([[lhs, rel]]).repeat(args.num_ent, axis=0), cst_inds
],
axis=1)
l_batch = ltensor(l_batch).contiguous()
r_batch = ltensor(r_batch).contiguous()
if args.use_cuda:
l_batch = l_batch.cuda()
r_batch = r_batch.cuda()
l_batch = Variable(l_batch)
r_batch = Variable(r_batch)
l_enrgs = modelD(l_batch)
r_enrgs = modelD(r_batch)
l_rank = compute_rank(v2np(l_enrgs), lhs)
r_rank = compute_rank(v2np(r_enrgs), rhs)
l_ranks.append(l_rank)
r_ranks.append(r_rank)
l_ranks = np.array(l_ranks)
r_ranks = np.array(r_ranks)
return l_ranks, r_ranks