In [ ]:

    

import tensorflow as tf

from gan import GAN
from wgan import WGAN
from cramer_gan import CramerGAN


    

In [ ]:

    

data_params = {
    'n_mixture': 8,
    'std': 0.05,
    'radius': 1,
}
generator_params = {
    'n_layers': 4,
    'n_hidden': 256,
    'activation_fn': tf.nn.relu,
}
discriminator_params = {
    'n_layers': 4,
    'n_hidden': 256,
    'activation_fn': tf.nn.relu,
}

iterations = 50000
batch_size = 128
save_step = 2500


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'modified_objective': False,
    'optimization': {
        'algorithm': 'alternating',
        'discriminator_steps': 5,
    },
}

tf.reset_default_graph()
gan = GAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='alternating-gan')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'modified_objective': True,
    'optimization': {
        'algorithm': 'alternating',
        'discriminator_steps': 5,
    },
}

tf.reset_default_graph()
gan = GAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='alternating-gan-2')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'gradient_penalty': False,
    'optimization': {
        'algorithm': 'alternating',
        'discriminator_steps': 5,
    },
}

tf.reset_default_graph()
gan = WGAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='alternating-wgan')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'gradient_penalty': True,
    'lambda': 0.1,
    'optimization': {
        'algorithm': 'alternating',
        'discriminator_steps': 5,
    },
}

tf.reset_default_graph()
gan = WGAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='alternating-wgan-gp')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': {
        'n_layers': 4,
        'n_hidden': 256,
        'activation_fn': tf.nn.relu,
        'output_dim': 256,
    },
    'lambda': 0.1,
    'optimization': {
        'algorithm': 'alternating',
        'discriminator_steps': 5,
    },
}

tf.reset_default_graph()
gan = CramerGAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='alternating-cramer-gan')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'modified_objective': False,
    'optimization': {
        'algorithm': 'consensus',
        'gamma': 10,
    },
}

tf.reset_default_graph()
gan = GAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='consensus-gan')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'modified_objective': True,
    'optimization': {
        'algorithm': 'consensus',
        'gamma': 10,
    },
}

tf.reset_default_graph()
gan = GAN(params)
gan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='consensus-gan-2')


    

In [ ]:

    

params = {
    'z_dim': 64,
    'data': data_params,
    'generator': generator_params,
    'discriminator': discriminator_params,
    'gradient_penalty': False,
    'optimization': {
        'algorithm': 'consensus',
        'gamma': 10,
    },
}

tf.reset_default_graph()
wgan = WGAN(params)
wgan.train(iterations=iterations, batch_size=batch_size, save_step=save_step, dirname='consensus-wgan')