Few-Shot and Zero-Shot Multi-Label Learning for Structured Label Spaces

Large multi-label datasets contain labels that occur thousands of times (frequent group), those that occur only a few times (few-shot group), and labels that never appear in the training dataset (zero-shot group). Multi-label few- and zero-shot label prediction is mostly unexplored on datasets with large label spaces, especially for text classification. In this repository, we have the code for our few- and zero-shot method for multilabel text classification when there is a known structure over the label space.

Note: Examples of the data format can be found in the "data" folder.

Required Packages

• Python 2.7
• numpy 1.11.1+
• scipy 0.18.0+
• Theano
• gensim
• sklearn
• nltk

Also, you will need to have a set of pretrained embeddings. You can point to your embeddings by changing line 105 of load_data.py

Usage

Training

python train.py --num_epochs 25 --word_vectors 'gensim_w2v_pubmed' --model_type cnn --train_data_X './data/train_data.json' --val_data_X './data/dev_data.json' --checkpoint_dir './checkpoints' --num_feat_maps 300 --grad_clip 3 --min_df 5 --lr 0.0001 --penalty 0.0000 --dropout 0.5 --lr_decay 0.0000 --cnn_conv_size 3 4 5  --checkpoint_name my_model_name

usage: train_match.py [-h] [--num_epochs NUM_EPOCHS] [--num_models NUM_MODELS]
                      [--word_vectors WORD_VECTORS] [--labels LABELS]
                      [--checkpoint_dir CHECKPOINT_DIR]
                      [--checkpoint_name CHECKPOINT_NAME]
                      [--hidden_state HIDDEN_STATE]
                      [--learn_embeddings LEARN_EMBEDDINGS] [--min_df MIN_DF]
                      [--lr LR] [--penalty PENALTY] [--dropout DROPOUT]
                      [--lr_decay LR_DECAY] [--minibatch_size MINIBATCH_SIZE]
                      [--val_minibatch_size VAL_MINIBATCH_SIZE]
                      [--model_type MODEL_TYPE] [--train_data_X TRAIN_DATA_X]
                      [--val_data_X VAL_DATA_X] [--seed SEED]
                      [--grad_clip GRAD_CLIP]
                      [--cnn_conv_size CNN_CONV_SIZE [CNN_CONV_SIZE ...]]
                      [--num_feat_maps NUM_FEAT_MAPS] [--num_att NUM_ATT]
                      [--num_support NUM_SUPPORT]

Train Neural Network.

optional arguments:
  -h, --help            show this help message and exit
  --num_epochs NUM_EPOCHS
                        Number of updates to make.
  --num_models NUM_MODELS
                        Number of updates to make.
  --word_vectors WORD_VECTORS
                        Word vecotors filepath.
  --labels LABELS       All Labels.
  --checkpoint_dir CHECKPOINT_DIR
                        Checkpoint directory.
  --checkpoint_name CHECKPOINT_NAME
                        Checkpoint File Name.
  --hidden_state HIDDEN_STATE
                        hidden layer size.
  --learn_embeddings LEARN_EMBEDDINGS
                        Learn Embedding Parameters.
  --min_df MIN_DF       Min word count.
  --lr LR               Learning Rate.
  --penalty PENALTY     Regularization Parameter.
  --dropout DROPOUT     Dropout Value.
  --lr_decay LR_DECAY   Learning Rate Decay.
  --minibatch_size MINIBATCH_SIZE
                        Mini-batch Size.
  --val_minibatch_size VAL_MINIBATCH_SIZE
                        Val Mini-batch Size.
  --model_type MODEL_TYPE
                        Neural Net Architecutre.
  --train_data_X TRAIN_DATA_X
                        Training Data.
  --val_data_X VAL_DATA_X
                        Validation Data.
  --seed SEED           Random Seed.
  --grad_clip GRAD_CLIP
                        Gradient Clip Value.
  --cnn_conv_size CNN_CONV_SIZE [CNN_CONV_SIZE ...]
                        CNN Covolution Sizes (widths)
  --num_feat_maps NUM_FEAT_MAPS
                        Number of CNN Feature Maps.
  --num_att NUM_ATT     Number of Heads.
  --num_support NUM_SUPPORT
                        Number nearest neighbors to sample for each input
                        instance.

Acknowledgements

Anthony Rios and Ramakanth Kavuluru. "Few-Shot and Zero-Shot Multi-Label Learning for Structured Label Spaces". EMNLP 2018

@inproceedings{arios2018emrzero,
  title={Few-Shot and Zero-Shot Multi-Label Learning for Structured Label Spaces},
  author={Rios, Anthony and Kavuluru, Ramakanth},
  booktitle={Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing},
  year={2018}
}

Written by Anthony Rios (anthonymrios at gmail dot com)