.
├── model/                      # The model implementation containing the sclstm layers  
│  ├── lm_deep.py               # SCLSTM basedimplementation of utterance generation
│  ├── cvae.py                  # SCVAE based implementation of utterance generation
│  ├── masked_cross_entropy.py  # Masked cross entropy loss
│  ├── masked_kl_divergence.py  # Masked KL-divergence loss
│  ├── model_util.py            # Some helper functions of model
│  ├── layers/                  # The unique domain data split
│  │   ├── sclstm.py            # SCLSTM network
│  │   ├── encoder.py           # Encoder for SCVAE
│  │   └── decoder_deep.py      # SCLSTM based decoder
├── loader                      # Dataset loader
│  ├── dataset_woz3.py          # Loader of original dataset
│  └── task.py                  # Loader of task and implementation of examplars
├── resource/                   # Data source
│  ├── data_split/*.json        # Data split file
│  ├── feat_*.json              # Feature file containing dialogue acts and slot-values
│  ├── text_*.txt               # Text file with original and delexicalized utterances
│  ├── vacab.txt                # Vocabulary
│  └── template.txt             # The template containing possible domain-action-slot-value
├── ewc.py                      # Compute ewc penalty
├── bleu.py                     # Compute bleu score
├── run_woz3.py                 # Code for starting training, evaluating and testing
├── run.sh                      # Script to run experiment
├── config/*.cfg                # Experiment configuration 
├── experiments/experiment_name # Model and results of experiments
│  ├── model/                   # Models throughout the training of task sequence, notice that the intermediate model.pt contains the test results on all tasks till current task
│  ├── examplars/               # Examplars of each task
│  ├── experiment_name.log      # Training log
│  └── experiment_name.res      # Testing result
├── view_results.ipynb          # Summarize statistics of experiments
└── 
...

.
├── split_multi_task.py                 # Script for generating utterances with unique task (domain/dialogue act), run this only if you need to re-generate feat.json, text.json and data split
├── construct_examplar.py               # Construct exemplars with `herding`, `random` or `prioritized (loss)` based method
└── 
...