0. Requirements

   • python3.+, Preferably python3.7

1. Create a virtual environment

   • python3 -m venv $env_name, you must specify $env_name
   • for example, python3 -m venv SeqVAE

2. Setting up the environment

   • Activate the environment, for example, use source ./bin/activate
   • Install dependencies, pip3 install -r requirements.txt

The research paper SeqVAE

1. Project Structure

   .
   ├── README.md
   ├── configs.py                  # some configuration
   ├── data.py                     # define how to parse and encode `midi`, from `MusicVAE`
   ├── decoder.py
   ├── encoder.py
   │── metric.py                   # QN SPB evaluation
   ├── interpolate.py
   ├── lstm_utils.py               # encapsulating the use of lstm, from `magenta`   
   ├── output
   │   ├── create
   │   ├── interpolate
   │   │   ├── ashover2.mid
   │   │   └── ashover7.mid
   │   ├── nottingham.tfrecord     # training set
   │   ├── nottingham_100.npy
   │   └── train                   # save the trained model
   │       └── seqvae_1-20200220-1422
   │           ├── checkpoint
   │           ├── model.ckpt-23000.data-00000-of-00001
   │           ├── model.ckpt-23000.index
   │           └── model.ckpt-23000.meta
   ├── requirements.txt
   ├── rnn_discriminator.py
   ├── seq_vae_generate.py
   ├── seq_vae_model.py
   ├── seq_vae_train.py
   └── utils.py
   

2. Use

   • how to train, python seq_vae_train.py ...;
   • how to generate, python seq_vae_generate.py ...;
   • how to interpolate, python interpolate.py ...;
   • ... means that you need to add some options. Option description in related python file.