def AIC_train():
# bootstrapping에 사용되는 학습데이터는 "PIC_train"을 통해 PIC 처리를 사전에 진행해야 함.
for epoch in range(1, config.boot_iter):
print("iter : ", config.iter)
main_tagger_AIC = Tagger()
main_tagger_AIC.main_trainAIC()
main_tagger_AIC.evaluateAIC("main_model")
self_trainig = main_tagger_AIC.main_taggingAIC(
mode="self_tagging") # init self training data\
# 배깅 모델 학습
splited_labels, splited_features, splited_sentences = split_self_labeling(
self_trainig[0], self_trainig[1], self_trainig[2])
print("%s_iter(main) -> self_labled_s1 : %s" %
(config.iter, len(self_trainig[0])))
for model_idx in range(1, config.model_num + 1):
print("model_idx : ", model_idx)
bagging_taggerAIC = Tagger()
bagging_taggerAIC.bagging_trainAIC("bagging_train", model_idx,
splited_features[model_idx - 1],
splited_labels[model_idx - 1])
print("bagging model%s acc" % model_idx)
bagging_taggerAIC.evaluateAIC("bagging_eval",
model_idx) # each bagging model
bagging_bootstrap(model_idx, splited_sentences[model_idx - 1],
splited_labels[model_idx - 1])
if model_idx == 1:
score_i, raw_sentences, features = bagging_taggerAIC.bagging_taggingAIC(
"self_tagging", model_idx)
else:
new_socre_i = bagging_taggerAIC.bagging_taggingAIC(
"self_tagging", model_idx)
for idx, _ in enumerate(new_socre_i):
score_i[idx] = np.asarray(score_i[idx]) + np.asarray(
new_socre_i[idx])
predicts = bagging_taggerAIC.score2tag(score_i, raw_sentences,
features)
main_bootstrap(predicts)
config.iter += 1
