def __pretrain_lstmcrf(word_vecs_file, pre_tok_texts_file, pre_aspect_terms_file, pre_opinion_terms_file,
dst_model_file, task):
init_logging('log/nr-pre-{}.log'.format(str_today), mode='a', to_stdout=True)
n_tags = 5 if task == 'both' else 3
print('loading data ...')
with open(word_vecs_file, 'rb') as f:
vocab, word_vecs_matrix = pickle.load(f)
load_model_file = None
save_model_file = dst_model_file
if task == 'both':
train_data, valid_data = datautils.get_data_amazon_ao(
vocab, pre_aspect_terms_file, pre_opinion_terms_file, pre_tok_texts_file)
elif task == 'aspect':
train_data, valid_data = datautils.get_data_amazon(
vocab, pre_aspect_terms_file, pre_tok_texts_file, task)
else:
train_data, valid_data = datautils.get_data_amazon(
vocab, pre_opinion_terms_file, pre_tok_texts_file, task)
print('done')
# lstmcrf = LSTMCRF(n_tags, word_vecs_matrix, hidden_size_lstm=hidden_size_lstm)
lstmcrf = LSTMCRF(n_tags, word_vecs_matrix, hidden_size_lstm=hidden_size_lstm, model_file=load_model_file)
# print(valid_data.aspects_true_list)
lstmcrf.train(train_data.word_idxs_list, train_data.labels_list, valid_data.word_idxs_list,
valid_data.labels_list, vocab, valid_data.tok_texts, valid_data.aspects_true_list,
valid_data.opinions_true_list, n_epochs=n_epochs, save_file=save_model_file)
def __pre_train_nrdj(word_vecs_file,
tok_texts_file,
aspect_terms_file,
opinion_terms_file,
dst_model_file,
task,
load_model_file=None):
init_logging('log/nrdj-pre-{}.log'.format(str_today),
mode='a',
to_stdout=True)
# n_train = 1000
n_train = -1
label_opinions = True
# label_opinions = False
n_tags = 5 if label_opinions else 3
# n_tags = 5 if task == 'train' else 3
batch_size = 20
lr = 0.001
share_lstm = False
print('loading data ...')
with open(word_vecs_file, 'rb') as f:
vocab, word_vecs_matrix = pickle.load(f)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE1_FILE)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE3_FILE)
train_data_src1, valid_data_src1 = datautils.get_data_amazon(
vocab, aspect_terms_file, tok_texts_file, 'aspect')
train_data_src2, valid_data_src2 = datautils.get_data_amazon(
vocab, opinion_terms_file, tok_texts_file, 'opinion')
print('done')
nrdj = NeuRuleDoubleJoint(n_tags,
word_vecs_matrix,
share_lstm,
hidden_size_lstm=hidden_size_lstm,
model_file=load_model_file,
batch_size=batch_size)
nrj_train_data_src1 = nrj_train_data_src2 = None
# if train_mode != 'target-only':
# nrj_train_data_src1 = NeuRuleDoubleJoint.TrainData(
# train_data_src1.word_idxs_list, train_data_src1.labels_list, valid_data_src1.word_idxs_list,
# valid_data_src1.labels_list, valid_data_src1.tok_texts, valid_data_src1.aspects_true_list, None
# )
# nrj_train_data_src2 = NeuRuleDoubleJoint.TrainData(
# train_data_src2.word_idxs_list, train_data_src2.labels_list, valid_data_src2.word_idxs_list,
# valid_data_src2.labels_list, valid_data_src2.tok_texts, None,
# valid_data_src2.opinions_true_list
# )
nrdj.pre_train(train_data_src1,
valid_data_src1,
train_data_src2,
valid_data_src2,
vocab,
n_epochs=30,
lr=lr,
save_file=dst_model_file)
def __train_dlc(word_vecs_file, train_tok_texts_file, train_sents_file, train_valid_split_file, test_tok_texts_file,
test_sents_file):
init_logging('log/dlc-jtrain2-{}.log'.format(str_today), mode='a', to_stdout=True)
# n_train = 1000
n_train = -1
label_opinions = True
# label_opinions = False
n_tags = 5 if label_opinions else 3
# n_tags = 5 if task == 'train' else 3
batch_size = 10
lr = 0.001
share_lstm = False
print('loading data ...')
with open(word_vecs_file, 'rb') as f:
vocab, word_vecs_matrix = pickle.load(f)
logging.info('word vec dim: {}, n_words={}'.format(word_vecs_matrix.shape[1], word_vecs_matrix.shape[0]))
train_data_src1, valid_data_src1 = datautils.get_data_amazon(
vocab, pre_aspect_terms_file, pre_tok_texts_file, 'aspect')
train_data_src2, valid_data_src2 = datautils.get_data_amazon(
vocab, pre_opinion_terms_file, pre_tok_texts_file, 'opinion')
train_data, valid_data, test_data = datautils.get_data_semeval(
train_sents_file, train_tok_texts_file, train_valid_split_file, test_sents_file, test_tok_texts_file,
vocab, n_train, label_opinions)
print('done')
dlc = DSLSTMCRF(word_vecs_matrix, hidden_size_lstm=hidden_size_lstm,
model_file=None, batch_size=batch_size)
dlc.joint_train(train_data_src1, valid_data_src1, train_data_src2, valid_data_src2,
train_data, valid_data, test_data, n_epochs=n_epochs, lr=lr)
def __pre_train_nrdj(word_vecs_file,
tok_texts_file,
aspect_terms_file,
opinion_terms_file,
dst_model_file,
task,
lamb,
lstm_l2,
train_word_embeddings=False,
load_model_file=None):
init_logging('log/{}-pre-{}-{}.log'.format(
os.path.splitext(os.path.basename(__file__))[0],
utils.get_machine_name(), str_today),
mode='a',
to_stdout=True)
# n_train = 1000
n_train = -1
label_opinions = True
# label_opinions = False
n_tags = 5 if label_opinions else 3
# n_tags = 5 if task == 'train' else 3
batch_size = 32
lr = 0.001
share_lstm = False
logging.info(word_vecs_file)
logging.info(aspect_terms_file)
logging.info(opinion_terms_file)
logging.info('dst: {}'.format(dst_model_file))
print('loading data ...')
vocab, word_vecs_matrix = __load_word_vecs(word_vecs_file)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE1_FILE)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE3_FILE)
train_data_src1, valid_data_src1 = datautils.get_data_amazon(
vocab, aspect_terms_file, tok_texts_file, 'aspect')
train_data_src2, valid_data_src2 = datautils.get_data_amazon(
vocab, opinion_terms_file, tok_texts_file, 'opinion')
print('done')
logging.info('train_word_embeddings={} lstm_l2={}'.format(
train_word_embeddings, lstm_l2))
nrdj = NeuRuleDoubleJoint(n_tags,
word_vecs_matrix,
share_lstm,
hidden_size_lstm=hidden_size_lstm,
train_word_embeddings=train_word_embeddings,
lamb=lamb,
lstm_l2_src=lstm_l2,
model_file=load_model_file,
batch_size=batch_size)
nrdj.pre_train(train_data_src1,
valid_data_src1,
train_data_src2,
valid_data_src2,
vocab,
n_epochs=50,
lr=lr,
save_file=dst_model_file)
def __train_neurule_double_joint():
init_logging('log/nrdj-{}.log'.format(str_today), mode='a', to_stdout=True)
# n_train = 1000
n_train = -1
# task = 'pretrain'
task = 'train'
label_opinions = True
n_tags = 5 if label_opinions else 3
# n_tags = 5 if task == 'train' else 3
batch_size = 20
lr = 0.001
share_lstm = False
train_mode = 'target-only'
print('loading data ...')
with open(config.SE14_LAPTOP_GLOVE_WORD_VEC_FILE, 'rb') as f:
vocab, word_vecs_matrix = pickle.load(f)
train_data_tar, valid_data_tar = datautils.get_data_semeval(
config.SE14_LAPTOP_TRAIN_SENTS_FILE, config.SE14_LAPTOP_TRAIN_TOK_TEXTS_FILE,
config.SE14_LAPTOP_TEST_SENTS_FILE, config.SE14_LAPTOP_TEST_TOK_TEXTS_FILE,
vocab, n_train, label_opinions)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE1_FILE)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE3_FILE)
train_data_src1, valid_data_src1 = datautils.get_data_amazon(
vocab, config.AMAZON_TERMS_TRUE2_FILE, config.AMAZON_TOK_TEXTS_FILE, 'aspect')
train_data_src2, valid_data_src2 = datautils.get_data_amazon(
vocab, config.AMAZON_TERMS_TRUE4_FILE, config.AMAZON_TOK_TEXTS_FILE, 'opinion')
rule_model_file = config.LAPTOP_NRDJ_RULE_MODEL_FILE if task == 'train' else None
# rule_model_file = None
pretrain_model_file = config.LAPTOP_NRDJ_RULE_MODEL_FILE
save_model_file = config.LAPTOP_NRDJ_RULE_MODEL_FILE
print('done')
nrdj = NeuRuleDoubleJoint(n_tags, word_vecs_matrix, share_lstm,
hidden_size_lstm=hidden_size_lstm,
model_file=rule_model_file)
nrj_train_data_src1 = nrj_train_data_src2 = None
# if train_mode != 'target-only':
nrj_train_data_src1 = NeuRuleDoubleJoint.TrainData(
train_data_src1.word_idxs_list, train_data_src1.labels_list, valid_data_src1.word_idxs_list,
valid_data_src1.labels_list, valid_data_src1.tok_texts, valid_data_src1.aspects_true_list, None
)
nrj_train_data_src2 = NeuRuleDoubleJoint.TrainData(
train_data_src2.word_idxs_list, train_data_src2.labels_list, valid_data_src2.word_idxs_list,
valid_data_src2.labels_list, valid_data_src2.tok_texts, None,
valid_data_src2.opinions_true_list
)
nrj_train_data_tar = NeuRuleDoubleJoint.TrainData(
train_data_tar.word_idxs_list, train_data_tar.labels_list, valid_data_tar.word_idxs_list,
valid_data_tar.labels_list, valid_data_tar.tok_texts, valid_data_tar.aspects_true_list,
valid_data_tar.opinions_true_list
)
if task == 'pretrain':
nrdj.pre_train(nrj_train_data_src1, nrj_train_data_src2, vocab, n_epochs=n_epochs, lr=lr,
save_file=pretrain_model_file)
if task == 'train':
nrdj.train(nrj_train_data_src1, nrj_train_data_src2, nrj_train_data_tar, vocab, train_mode,
n_epochs=n_epochs, lr=lr)
def __train_nrdj_deep_restaurant_pr():
init_logging('log/nrdj-deep-restaurant-{}.log'.format(str_today),
mode='a',
to_stdout=True)
# n_train = 1000
n_train = -1
task = 'train'
label_task = 'aspect'
n_tags = 5 if label_task == 'both' else 3
# n_tags = 5 if task == 'train' else 3
batch_size = 20
hidden_size_lstm = 100
n_epochs = 500
lr = 0.001
share_lstm = True
# load_pretrained_model = True
load_pretrained_model = False
# train_mode = 'target-only'
train_mode = 'all'
aspect_terms_p_file = 'd:/data/aspect/semeval14/restaurant/yelp-aspect-rule-result-p.txt'
aspect_terms_r_file = 'd:/data/aspect/semeval14/restaurant/yelp-aspect-rule-result-r.txt'
# opinion_terms_file = 'd:/data/aspect/semeval14/restaurant/yelp-opinion-rule-result.txt'
yelp_tok_texts_file = 'd:/data/res/yelp-review-eng-tok-sents-round-9.txt'
rule_model_file = 'd:/data/aspect/semeval14/tf-model/drest/yelp-nrdj.ckpl'
# rule_model_file = None
load_model_file = None
if task == 'train' and load_pretrained_model:
load_model_file = rule_model_file
# save_model_file = None if task == 'train' else rule_model_file
save_model_file = rule_model_file if task == 'pretrain' else None
print('loading data ...')
with open(config.SE14_REST_GLOVE_WORD_VEC_FILE, 'rb') as f:
vocab, word_vecs_matrix = pickle.load(f)
train_data_tar, valid_data_tar = datautils.get_data_semeval(
config.SE14_REST_TRAIN_SENTS_FILE,
config.SE14_REST_TRAIN_TOK_TEXTS_FILE,
config.SE14_REST_TEST_SENTS_FILE, config.SE14_REST_TEST_TOK_TEXTS_FILE,
vocab, n_train, label_task)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE1_FILE)
# train_data_src1, valid_data_src1 = __get_data_amazon(vocab, config.AMAZON_TERMS_TRUE3_FILE)
train_data_src1, valid_data_src1 = datautils.get_data_amazon(
vocab, aspect_terms_p_file, yelp_tok_texts_file, 'aspect')
# train_data_src2, valid_data_src2 = datautils.get_data_amazon(
# vocab, aspect_terms_r_file, yelp_tok_texts_file, 'opinion')
train_data_src2, valid_data_src2 = datautils.get_data_amazon(
vocab, aspect_terms_r_file, yelp_tok_texts_file, 'aspect')
# train_data_src2, valid_data_src2 = datautils.get_data_amazon(
# vocab, opinion_terms_file, yelp_tok_texts_file, 'opinion')
print('done')
nrdj = NeuRuleDoubleJointDeep(n_tags,
word_vecs_matrix,
share_lstm,
hidden_size_lstm=hidden_size_lstm,
model_file=load_model_file)
nrj_train_data_src1 = nrj_train_data_src2 = None
# if train_mode != 'target-only':
nrj_train_data_src1 = NeuRuleDoubleJointDeep.TrainData(
train_data_src1.word_idxs_list, train_data_src1.labels_list,
valid_data_src1.word_idxs_list, valid_data_src1.labels_list,
valid_data_src1.tok_texts, valid_data_src1.aspects_true_list, None)
nrj_train_data_src2 = NeuRuleDoubleJointDeep.TrainData(
train_data_src2.word_idxs_list, train_data_src2.labels_list,
valid_data_src2.word_idxs_list, valid_data_src2.labels_list,
valid_data_src2.tok_texts, valid_data_src2.aspects_true_list, None)
nrj_train_data_tar = NeuRuleDoubleJointDeep.TrainData(
train_data_tar.word_idxs_list, train_data_tar.labels_list,
valid_data_tar.word_idxs_list, valid_data_tar.labels_list,
valid_data_tar.tok_texts, valid_data_tar.aspects_true_list,
valid_data_tar.opinions_true_list)
nrdj.train(nrj_train_data_src1,
nrj_train_data_src2,
nrj_train_data_tar,
vocab,
train_mode,
n_epochs=n_epochs,
lr=lr)