Technical Report: Relational Data Synthesis using Generative Adversarial Networks: A Design Space Exploration
https://github.com/ruclty/Daisy/blob/master/daisy.pdf
Before running the code, please make sure your Python version is above 3.6. Then install necessary packages by :
pip install -r requirements.txtAll public datasets we used in our work can be downloaded from the datasets page.
You need to write a .json file as the configuration. The keyworks should include :
- name: required, name of the output file
- train: required, path of the training file
- sample: required, path of the sampling file
- normalize_cols: required, a list contains index of the normalize columns
- gmm_cols: required, a list contains index of the gmm columns
- one-hot_cols: required, a list contains index of the one-hot columns
- ordinal_cols: required, a list contains index of the ordinal columns
- model: required, model of the generator, LGAN(LSTM) or VGAN(MLP)
- dis_model: optional, model of the discriminator, lstm or mlp, default mlp
- n_epochs: required, num of training epochs
- steps_per_epoch: required, steps per epoch
- n_search: required, training times
- rand_search: required, whether to search hyper-parameters randomly
- param: required if rand_search is 'no', hyper-parameter of the NN
- train_method: required, training method
- label: required if train_method is a conditional training, name of the label column
- KL: optional, whether using KL loss in training, default 'yes'
- ratio: optional, the ratio of the number of sample records to the real data, default 1
- sample_times: optional, times of sampling, default 1
Folder "params" contains some examples, you can run the code using those parameter files directly, or write a self-defined parameter file to train a new dataset.
Run the code with the following command :
python Daisy/run.py [parameter file]or run the following command for quickly start :
./run.sh