text-classification-tensorflow
Tensorflow implementation of Attention-based Bidirectional RNN text classification. (with naive Bidirectional RNN model as a baseline)
Requirements
- Python 3
- Tensorflow
- pip install -r requirements.txt
Usage
Prepare Data
We are using pre-processed version of Twitter Sentiment Classification Data. To use sample data (100K train/30K test),
$ unzip sample_data/sample_data.zip -d sample_data
To use full data (1.2M train/0.4M test), download it from google drive link.
To use Glove pre-trained embedding, download it via
$ python download_glove.py
Train
To train the model with sample data,
$ python train.py
Train data is split to train set(85%) and validation set(15%). Every 2000 steps, the classification accuracy is tested with validation set and the best model is saved.
To use Glove pre-trained vectors as initial embedding,
$ python train.py --glove
Additional Hyperparameters
$ python train.py -h
usage: train.py [-h] [--train_tsv TRAIN_TSV] [--model MODEL] [--glove]
[--embedding_size EMBEDDING_SIZE] [--num_hidden NUM_HIDDEN]
[--num_layers NUM_LAYERS] [--learning_rate LEARNING_RATE]
[--batch_size BATCH_SIZE] [--num_epochs NUM_EPOCHS]
[--keep_prob KEEP_PROB] [--checkpoint_dir CHECKPOINT_DIR]
optional arguments:
-h, --help show this help message and exit
--train_tsv TRAIN_TSV
Train tsv file.
--model MODEL naive | att
--glove Use glove as initial word embedding.
--embedding_size EMBEDDING_SIZE
Word embedding size. (For glove, use 50 | 100 | 200 | 300)
--num_hidden NUM_HIDDEN
RNN Network size.
--num_layers NUM_LAYERS
RNN Network depth.
--learning_rate LEARNING_RATE
Learning rate.
--batch_size BATCH_SIZE
Batch size.
--num_epochs NUM_EPOCHS
Number of epochs.
--keep_prob KEEP_PROB
Dropout keep prob.
--checkpoint_dir CHECKPOINT_DIR
Checkpoint directory.
Test
To test classification accuracy for test data,
$ python test.py
To use custom data,
$ python test.py --test_tsv=<CUSTOM_TSV>
Sample Test Results
Trained and tested with full data with default hyper-parameters,
| Model | Naive | Naive(+Glove) | Attention | Attention(+Glove) |
|---|---|---|---|---|
| Accuracy | 0.574 | 0.578 | 0.811 | 0.820 |