WGAN_GP synthetic fraud

WGAN_synth.py

from ydata_synthetic.synthesizers.regular import WGAN_GP

noise_dim = 32
dim = 128
batch_size = 164

log_step = 100
epochs = 1000+1
learning_rate = 5e-4
beta_1 = 0.5
beta_2 = 0.9
models_dir = './cache'

gan_args = [batch_size, learning_rate, beta_1, beta_2, noise_dim, fraud.shape[1]-1, dim]
train_args = ['', epochs, log_step]

fraud_synth = WGAN_GP(gan_args, n_critic=2)
fraud_synth.train(fraud.drop('y', axis=1), train_args)

synthetic_fraud = fraud_synth.sample(400)
synthetic_fraud.columns = fraud.drop('y', axis=1).columns