def eval_test(self):
"""do the evaluation when training
:return:
"""
tvars = tf.trainable_variables()
tvars_vals = self.sess.run(tvars)
#for var, val in zip(tvars, tvars_vals):
# print(var)
results = []
#range_=[4,5]
if config.app == "citeseerNew":
for i in range(2):
LPE = elp.LinkPredictEval(
config.emb_filenames[i + 4], config.test_filename,
config.test_neg_filename, config.n_node, config.n_embed,
config.n_features, config.pretrain_feature_filename,
config.emb_filenames[i + 6], config.modes[i])
result = LPE.eval_link_prediction()
#results.append(config.modes[i] + ":" + str(result) + "\n")
results.append(config.modes[i] + ":" + "median--" +
str(result[0]) + "--" + str(result[1]) +
" softmax--" + str(result[2]) + "\n")
if (float(result[2]) >= self.maxSoftmaxGen and i == 0):
print(self.maxSoftmaxGen)
for var, val in zip(tvars, tvars_vals):
if "generator/Generator/Variable_1:0" in var.name:
np.savetxt(config.emb_filenames[10], val)
self.maxSoftmaxGen = float(result[2])
self.write_emb_to_txt(0, 2)
if (float(result[2]) >= self.maxSoftmaxDis and i == 1):
print(self.maxSoftmaxDis)
for var, val in zip(tvars, tvars_vals):
if "discriminator/discriminator/Variable_1:0" in var.name:
np.savetxt(config.emb_filenames[11], val)
self.maxSoftmaxDis = float(result[2])
self.write_emb_to_txt(1, 3)
if (float(result[1]) >= self.maxMedianGen and i == 0):
print(self.maxMedianGen)
for var, val in zip(tvars, tvars_vals):
if "generator/Generator/Variable_1:0" in var.name:
np.savetxt(config.emb_filenames[8], val)
self.maxMedianGen = float(result[1])
self.write_emb_to_txt(0, 0)
if (float(result[1]) >= self.maxMedianDis and i == 1):
print(self.maxMedianDis)
for var, val in zip(tvars, tvars_vals):
if "discriminator/discriminator/Variable_1:0" in var.name:
np.savetxt(config.emb_filenames[9], val)
self.maxMedianDis = float(result[1])
self.write_emb_to_txt(1, 1)
with open(config.result_filename, mode="a+") as f:
f.writelines(results)
def eval_test(self):
"""do the evaluation when training
:return:
"""
results = []
if config.app == "link_prediction":
for i in range(2):
LPE = elp.LinkPredictEval(config.emb_filenames[i],
config.test_filename,
config.test_neg_filename,
config.n_node, config.n_embed)
result = LPE.eval_link_prediction()
results.append(config.modes[i] + ":" + str(result) + "\n")
with open(config.result_filename, mode="a+") as f:
f.writelines(results)
def eval_test(config):
"""do the evaluation when training
:return:
"""
results = []
if config.app == "link_prediction":
LPE = elp.LinkPredictEval(config.emb_filename, config.test_filename,
config.test_neg_filename, config.n_node,
config.n_embed)
result = LPE.eval_link_prediction()
results.append(config.model + ":" + str(result) + "\n")
with open(config.result_filename, mode="a+") as f:
f.writelines(results)
test_edges = utils.read_edges_from_file(config.test_filename)
test_edges_neg = utils.read_edges_from_file(config.test_neg_filename)
test_edges.extend(test_edges_neg)
emd = utils.read_emd(config.emb_filename,
n_node=config.n_node,
n_embed=config.n_embed)
score_res = []
for i in range(len(test_edges)):
score_res.append(np.dot(emd[test_edges[i][0]], emd[test_edges[i][1]]))
test_label = np.array(score_res)
bar = np.median(test_label)
ind_pos = test_label >= bar
ind_neg = test_label < bar
test_label[ind_pos] = 1
test_label[ind_neg] = 0
true_label = np.zeros(test_label.shape)
true_label[0:len(true_label) // 2] = 1
f1 = f1_score(true_label, test_label, average='macro')
result = config.model + ":" + str(f1) + "\n"
print(result)
with open(config.result_filename_f1, mode="a+") as f:
f.writelines(result)