Requirements

• Tensorflow == 1.9.0
• NLTK
• joblib == 0.13.2

Files

• run_model.py: This file has all the flags needed for training, it chooses the data module, models and starts training.
• utils.py: Utilities module containing DataGenerators, Parallel model etc.
  • DataGenerator: Creates batches for vanilla NSE.
  • DataGeneratorHier: Creates batches for hierarchical NSE.
• Memory: All the files below contain a variant of NSE.
  • NSE: This is a neural semantic encoder class.
  • HierNSE: This is the hier-NSE class.
• models: All the files below contain an encoder, decoder, loss, optimizer functions that use an NSE.
  • model.py: Model using vanilla NSE.
  • model_hier.py: Model using hier-NSE (use this).
  • model_hier_sc.py: Self-Critic model (use this). It carefully back-propagates through the same multinomial samples that are sampled while forward pass.
• rouge: Rouge scripts used.
  • rouge_batch: A NumPy implementation (faster than existing ones). Used outside the TensorFlow graph.
• Data
  • Create a folder named data
  • Download the following splits into data folder: train, val, test
  • Download the CNN and Daily-Mail tokenized data: CNN, DM
  • Download GloVe

supervised model

• Training

    python run_model.py --model="hier" --mode="train" --PathToCheckpoint=/path/to/checkpoint --PathToTB=/path/to/tensorboard/logs
  

• Testing

  • Check the epoch number of the best supervised model from TensorBoard, let it be X

    python run_model.py --model="hier" --mode="test" --PathToCheckpoint=/path/to/checkpoint/model_epochX --PathToResults=/path/to/results
    

• Evaluation

    python run_model.py --model="hier" --mode="eval" --PathToResults=/path/to/results
  

self-critical model

• Training

  • First copy the best supervised model to the rl checkpoint.

      python run_model.py --model="rlhier" --mode="train" --restore=True --PathToCheckpoint=/path/to/checkpoint --PathToTB=/path/to/tensorboard/logs
    

• Testing

  • Check the epoch number of the best supervised model from TensorBoard, let it be X.

      python run_model.py --model="rlhier" --mode="train" --restore=True --PathToCheckpoint=/path/to/checkpoint/model_epochX --PathToResults=/path/to/results
    

• Evaluation

      python run_model.py --model="rlhier" --mode="eval" --PathToResults=/path/to/results