def test(self, save=True):
embedding = self.sess.run(self.output)
embeds1 = np.array([embedding[e] for e in self.kgs.test_entities1])
embeds2 = np.array([embedding[e] for e in self.kgs.test_entities2])
rest_12, _, _ = test(embeds1,
embeds2,
None,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=0,
accurate=True)
test(embeds1,
embeds2,
None,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=self.args.csls,
accurate=True)
if save:
ent_ids_rest_12 = [(self.kgs.test_entities1[i],
self.kgs.test_entities2[j])
for i, j in rest_12]
rd.save_results(self.out_folder, ent_ids_rest_12)
def test(self, save=True):
if self.args.test_method == "sa":
beta = self.args.beta
embeddings = np.concatenate(
[self.vec_se * beta, self.vec_ae * (1.0 - beta)], axis=1)
else:
embeddings = self.vec_se
embeds1 = np.array([embeddings[e] for e in self.kgs.test_entities1])
embeds2 = np.array([embeddings[e] for e in self.kgs.test_entities2])
rest_12, _, _ = test(embeds1,
embeds2,
None,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=0,
accurate=True)
test(embeds1,
embeds2,
None,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=self.args.csls,
accurate=True)
if save:
rd.save_results(self.out_folder, rest_12)
def test(self, save=True):
embeds1, embeds2, mapping = self._eval_test_embeddings()
rest_12, _, _ = test(embeds1,
embeds2,
mapping,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=0,
accurate=True)
test(embeds1,
embeds2,
mapping,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=self.args.csls,
accurate=True)
if save:
ent_ids_rest_12 = [(self.kgs.test_entities1[i],
self.kgs.test_entities2[j])
for i, j in rest_12]
rd.save_results(self.out_folder, ent_ids_rest_12)
def test_desc(self):
print("test desc")
test_links = pd.DataFrame(self.kgs.test_links)
desc1 = self.e_desc.loc[test_links.values[:, 0]].values
desc2 = self.e_desc.loc[test_links.values[:, 1]].values
desc_em1 = self.word_em[desc1]
desc_em2 = self.word_em[desc2]
dem1, dem2 = self.session.run(fetches=[self.desc_embedding1, self.desc_embedding2],
feed_dict={self.desc1: desc_em1,
self.desc2: desc_em2})
test(dem1, dem2, None, self.args.top_k,
self.args.test_threads_num, metric=self.args.eval_metric,
normalize=self.args.eval_norm, csls_k=self.args.csls, accurate=True)
def test(self, save=True):
embeds1, embeds2, mapping = self._eval_test_embeddings()
rest_12, _, _ = test(embeds1,
embeds2,
mapping,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=0,
accurate=True)
test(embeds1,
embeds2,
mapping,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
normalize=self.args.eval_norm,
csls_k=self.args.csls,
accurate=True)
if save:
rd.save_results(self.out_folder, rest_12)
def test(self, save=True):
mat = get_ent_embeds_from_attributes(self.kgs,
self.eval_attribute_embeddings(),
self.selected_attributes)
embeds1 = mat[self.kgs.test_entities1, ]
embeds2 = mat[self.kgs.test_entities2, ]
rest_12, _, _, rest_21, _, _ = evaluation.test(
embeds1,
embeds2,
None,
self.args.top_k,
self.args.test_threads_num,
metric=self.args.eval_metric,
csls_k=self.args.csls)
if save:
read.save_results(self.out_folder, rest_12)
def retest(self):
dir = self.out_folder.split("/")
new_dir = ""
for i in range(len(dir) - 2):
new_dir += (dir[i] + "/")
exist_file = os.listdir(new_dir)
new_dir = new_dir + exist_file[0] + "/"
embeds = np.load(new_dir + "ent_embeds.npy")
embeds1 = embeds[self.kgs.test_entities1]
embeds2 = embeds[self.kgs.test_entities2]
mapping = None
print(self.__class__.__name__, type(self.__class__.__name__))
if self.__class__.__name__ == "GCN_Align":
print(self.__class__.__name__, "loads attr embeds")
attr_embeds = np.load(new_dir + "attr_embeds.npy")
attr_embeds1 = attr_embeds[self.kgs.test_entities1]
attr_embeds2 = attr_embeds[self.kgs.test_entities2]
embeds1 = np.concatenate([embeds1 * self.args.beta, attr_embeds1 * (1.0 - self.args.beta)], axis=1)
embeds2 = np.concatenate([embeds2 * self.args.beta, attr_embeds2 * (1.0 - self.args.beta)], axis=1)
# if self.__class__.__name__ == "MTransE" or self.__class__.__name__ == "SEA" or self.__class__.__name__ == "KDCoE":
if os.path.exists(new_dir + "mapping_mat.npy"):
print(self.__class__.__name__, "loads mapping mat")
mapping = np.load(new_dir + "mapping_mat.npy")
print("conventional test:")
test(embeds1, embeds2, mapping, self.args.top_k, self.args.test_threads_num,
metric=self.args.eval_metric, normalize=self.args.eval_norm, csls_k=0, accurate=True)
print("conventional reversed test:")
if mapping is not None:
embeds1 = np.matmul(embeds1, mapping)
test(embeds2, embeds1, None, self.args.top_k, self.args.test_threads_num,
metric=self.args.eval_metric, normalize=self.args.eval_norm, csls_k=0, accurate=True)
else:
test(embeds2, embeds1, mapping, self.args.top_k, self.args.test_threads_num,
metric=self.args.eval_metric, normalize=self.args.eval_norm, csls_k=0, accurate=True)
print("stable test:")
stable_alignment(embeds1, embeds2, self.args.eval_metric, self.args.eval_norm, csls_k=0,
nums_threads=self.args.test_threads_num)
print("stable test with csls:")
stable_alignment(embeds1, embeds2, self.args.eval_metric, self.args.eval_norm, csls_k=self.args.csls,
nums_threads=self.args.test_threads_num)
