def on_epoch_end(self, epoch, logs=None):
print("pridict_dev")
pred_dev_y = model.predict(
[dev_x_indices, dev_x_segments, dev_entity1_pos, dev_entity2_pos],
batch_size=batch_size,
verbose=0)
# best_threshold, best_scores = threshold_search(y_develop,pred_dev_y)
precision, recall, best_score = f1(dev_y,
(pred_dev_y > 0.5).astype(int))
print(
"epoch###################################################################################################"
)
print("- dev F1 Score: {:.4f}".format(best_score))
logging.info("- dev F1 Score: " + str(best_score))
pred_test_y = model.predict([
test_x_indices, test_x_segments, test_entity1_pos, test_entity2_pos
],
batch_size=batch_size,
verbose=0)
p, r, f = f1(test_y, (pred_test_y > 0.5).astype(int))
print("- Val F1 Score: {:.4f}".format(f))
logging.info("- Val p Score:" + str(p))
logging.info("- Val r Score:" + str(r))
logging.info("- Val F1 Score:" + str(f))
if f > 0.6:
save_file("bert_hua_cea_" + str(f) + "_predict", pred_test_y)
model.save('./models/bert_hua_cea_' + str(f) + '.h5')
def post_process(predict_file):
test_entity = load_pkl('test_doc_entity_list')
true_y=load_pkl('tests_y')
file = open('../result/' + predict_file, 'rb')
predict = pickle.load(file)
file.close()
test = load_pkl('test_doc_entity')
y,y_dict=ansisy_predict(predict,test_entity)
doc_dict=ansis_doc(test)
new_y = match(y,y_dict,test_entity,doc_dict)
# p, r, f = f1(true_y, (predict > 0.5).astype(int))
p1, r1, f_1 = f1(true_y, new_y)
# print(f)
print(f_1)
print("///////")
return new_y,true_y
def post2(predict_file, delete_list,true_y,predict):
# true_y = load_pkl('tests_y')
# file = open('../result/' + predict_file, 'rb')
# predict = pickle.load(file)
# file.close()
new_pre=[]
new_y=[]
index=0
for i in range(len(predict)):
if i not in delete_list:
new_pre.append(predict[i])
new_y.append(true_y[i])
new_pre=np.array(new_pre)
new_y = np.array(new_y)
# p, r, f = f1(true_y, (predict > 0.5).astype(int))
p1, r1, f_1 = f1(new_y, (new_pre > 0.5).astype(int))
# print(f)
print(f_1)
intre_predict.append(1)
else:
intre_predict.append(0)
true_y = true_y.tolist()
for i in range(len(true_y)):
if i in list:
temp = true_y[i][0]
intre_true_y.append(true_y[i][0])
return intre_predict, intre_true_y
if __name__ == "__main__":
test_inter_sen_list = load_pkl('test_delete_list')
list = test_inter_sen_list[0]
file = open('../result/' + '0.5632286995515694_predict', 'rb')
predict = pickle.load(file)
file.close()
true_y = load_pkl('tests_y')
intra_predict, intra_true_y = intra_sen_classify(list, predict, true_y)
intre_predict, intre_true_y = inter_sen_classify(list, predict, true_y)
p1, r1, f_1 = f1(intra_true_y, intra_predict)
print(f_1)
p2, r2, f_2 = f1(intre_true_y, intre_predict)
print(f_2)
print()
