Skip to content

Latest commit

 

History

History
202 lines (161 loc) · 8 KB

README.md

File metadata and controls

202 lines (161 loc) · 8 KB
Raw

Template Code: BERT-for-Sequence-Labeling-and-Text-Classification

BERT is used for sequence annotation and text categorization template code to facilitate BERT for more tasks. The code has been tested on snips (intention recognition and slot filling task), ATIS (intention recognition and slot filling task) and conll-2003 (named entity recognition task) datasets. Welcome to use this BERT template to solve more NLP tasks, and then share your results and code here.

这是使用BERT进行序列标注和文本分类的模板代码，方便大家将BERT用于更多任务。该代码已经在SNIPS（意图识别和槽填充任务）、ATIS（意图识别和槽填充任务）和conll-2003（命名实体识别任务）数据集上进行了实验。欢迎使用这个BERT模板解决更多NLP任务，然后在这里分享你的结果和代码。

Task and Dataset

I have downloaded the data for you. Welcome to add new data set.

task name dataset name data source
CoNLL-2003 named entity recognition conll2003ner https://www.clips.uantwerpen.be/conll2003/ner/
Atis Joint Slot Filling and Intent Prediction atis https://github.com/MiuLab/SlotGated-SLU/tree/master/data/atis
Snips Joint Slot Filling and Intent Prediction snips https://github.com/MiuLab/SlotGated-SLU/tree/master/data/snips

Environment Requirements

Use pip install -r requirements.txt to install dependencies quickly.

  • python 3.6+
  • Tensorflow 1.12.0+
  • sklearn

Template Code Usage Method

Using pre training and fine-tuning model directly

For example: Atis Joint Slot Filling and Intent Prediction

  1. Download model weight atis_join_task_LSTM_epoch30_simple.zip and unzip then to file store_fine_tuned_model, https://pan.baidu.com/s/1SZkQXP8NrOtZKVEMfDE4bw;
  2. Run Code! You can change task_name and output_dir.
python run_slot_intent_join_task_LSTM.py \
  --task_name=Atis \
  --do_predict=true \
  --data_dir=data/atis_Intent_Detection_and_Slot_Filling \
  --vocab_file=pretrained_model/uncased_L-12_H-768_A-12/vocab.txt \
  --bert_config_file=pretrained_model/uncased_L-12_H-768_A-12/bert_config.json \
  --init_checkpoint=store_fine_tuned_model/atis_join_task_LSTM_epoch30_simple/model.ckpt-4198 \
  --max_seq_length=128 \
  --output_dir=./output_model_predict/atis_join_task_LSTM_epoch30_simple_ckpt4198

You can find the file of model prediction and the score of model prediction in output_dir (You can find the content of model socres later).

Quick start(model train and predict)

See predefined_task_usage.md for more predefined task usage codes.

  1. Move google's BERT code to file bert (I've prepared a copy for you.);
  2. Download google's BERT pretrained model and unzip then to file pretrained_model, https://github.com/google-research/bert;
  3. Run Code! You can change task_name and output_dir.

model training

python run_sequence_labeling_and_text_classification.py \
  --task_name=snips \
  --do_train=true \
  --do_eval=true \
  --data_dir=data/snips_Intent_Detection_and_Slot_Filling \
  --vocab_file=pretrained_model/uncased_L-12_H-768_A-12/vocab.txt \
  --bert_config_file=pretrained_model/uncased_L-12_H-768_A-12/bert_config.json \
  --init_checkpoint=pretrained_model/uncased_L-12_H-768_A-12/bert_model.ckpt \
  --num_train_epochs=3.0 \
  --output_dir=./store_fine_tuned_model/snips_join_task_epoch3/

Then you can find the fine tuned model in the output_dir=./store_fine_tuned_model/snips_join_task_epoch3/ folder.

model prediction

python run_sequence_labeling_and_text_classification.py \
  --task_name=Snips \
  --do_predict=true \
  --data_dir=data/snips_Intent_Detection_and_Slot_Filling \
  --vocab_file=pretrained_model/uncased_L-12_H-768_A-12/vocab.txt \
  --bert_config_file=pretrained_model/uncased_L-12_H-768_A-12/bert_config.json \
  --init_checkpoint=output_model/snips_join_task_epoch3/model.ckpt-1000 \
  --max_seq_length=128 \
  --output_dir=./output_model_prediction/snips_join_task_epoch3_ckpt1000

Then you can find the predicted output of the model and the output test results (accuracy, recall, F1 value, etc.) in the output_dir=./output_model_prediction/snips_join_task_epoch3_ckpt1000 folder.

File Structure

name function
bert store google's BERT code
data store task raw data set
output_model_prediction store model predict
store_fine_tuned_model store finet tuned model
calculating_model_score
pretrained_model store BERT pretrained model
run_sequence_labeling.py for Sequence Labeling Task
run_text_classification.py for Text Classification Task
run_sequence_labeling_and_text_classification.py for join task
calculate_model_score.py for evaluation model

Model Socres

The following model scores are model scores without careful adjustment of model parameters, that is to say, the scores can continue to improve!

CoNLL-2003 named entity recognition

eval_f = 0.926 eval_precision = 0.925 eval_recall = 0.928

Atis Joint Slot Filling and Intent Prediction

Intent Prediction Correct rate: 0.976 Accuracy: 0.976 Recall rate: 0.976 F1-score: 0.976

Slot Filling19 Correct rate: 0.955 Accuracy: 0.955 Recall rate: 0.955 F1-score: 0.955

How to add a new task

Just write a small piece of code according to the existing template!

Data

For example, If you have a new classification task QQP.

Before running this example you must download the GLUE data by running this script.

Code

Now, write code!

class QqpProcessor(DataProcessor):
    """Processor for the QQP data set."""

    def get_train_examples(self, data_dir):
        """See base class."""
        return self._create_examples(
            self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")

    def get_dev_examples(self, data_dir):
        """See base class."""
        return self._create_examples(
            self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")

    def get_test_examples(self, data_dir):
        """See base class."""
        return self._create_examples(
            self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")

    def get_labels(self):
        """See base class."""
        return ["0", "1"]

    def _create_examples(self, lines, set_type):
        """Creates examples for the training and dev sets."""
        examples = []
        for (i, line) in enumerate(lines):
            if i == 0 or len(line)!=6:
                continue
            guid = "%s-%s" % (set_type, i)
            text_a = tokenization.convert_to_unicode(line[3])
            text_b = tokenization.convert_to_unicode(line[4])
            if set_type == "test":
                label = "1"
            else:
                label = tokenization.convert_to_unicode(line[5])
            examples.append(
                InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
        return examples

Registration task

def main(_):
   tf.logging.set_verbosity(tf.logging.INFO)
   processors = {
       "qqp": QqpProcessor,
   }

Run

python run_text_classification.py \
--task_name=qqp \
--do_train=true \
--do_eval=true \
--data_dir=data/snips_Intent_Detection_and_Slot_Filling \
--vocab_file=pretrained_model/uncased_L-12_H-768_A-12/vocab.txt \
--bert_config_file=pretrained_model/uncased_L-12_H-768_A-12/bert_config.json \
--init_checkpoint=pretrained_model/uncased_L-12_H-768_A-12/bert_model.ckpt \
--max_seq_length=128 \
--train_batch_size=32 \
--learning_rate=2e-5 \
--num_train_epochs=3.0 \
--output_dir=./output/qqp_Intent_Detection/