WGAN experiment with MNIST

9fromlabmlimportexperiment1011fromlabml.configsimportcalculate

Import configurations from DCGAN experiment

13fromlabml\_nn.gan.dcganimportConfigs

Import Wasserstein GAN losses

16fromlabml\_nn.gan.wassersteinimportGeneratorLoss,DiscriminatorLoss

Set configurations options for Wasserstein GAN losses

19calculate(Configs.generator\_loss,'wasserstein',lambdac:GeneratorLoss())20calculate(Configs.discriminator\_loss,'wasserstein',lambdac:DiscriminatorLoss())

23defmain():

Create configs object

25conf=Configs()

Create experiment

27experiment.create(name='mnist\_wassertein\_dcgan',comment='test')

Override configurations

29experiment.configs(conf,30{31'discriminator':'cnn',32'generator':'cnn',33'label\_smoothing':0.01,34'generator\_loss':'wasserstein',35'discriminator\_loss':'wasserstein',36})

Start the experiment and run training loop

39withexperiment.start():40conf.run()414243if\_\_name\_\_=='\_\_main\_\_':44main()

labml.ai