- python 3.6
- pytorch 0.4.0
- nlgeval
conda install mkl-service
All the data can be found in the ofiicial repository here.
You need to first make symbolic link under data folder like this.
data/
├── word_embedding
├── textdata
├── cache
└── Audio-Visual-Feature
word_embedding/
└── glove.6B.300d.txt
python -m src.train --modelType [SimpleModel|....]
For more information, use --help flag.
A simple example can be found in train_simple.sh
python -m src.predict [parameter_timestep]
For more information, use --help flag.
.
├── data (all data should be placed here)
│ ├── Audio-Visual-Feature
│ ├── cache
│ ├── textdata
│ └── word_embedding
├── output (all model output would be place here)
│ ├── generate (after predicting, model prediced result will be place here as a JSON file)
│ ├── log (after training, training losses and validation losses will be placed here)
│ ├── metrics (after predicting, automatic evaluation metrics would be placed here)
│ ├── parameter (after training, model parameters would be placed here)
│ └── visualization
└── src
├── components (Layers that can be reuse)
├── dataset (handle the dataset)
├── log (log configuration file)
├── model (all models are here)
│ ├── components
│ └── net
├── statistics (plot things or compute statistics information)
└── util