return fig
N_SAMPLES = 1000
train_data = normal(size=(1, N_SAMPLES))
train_data = train_data.T
# TRAIN KDE #
kde_model = KDE(use_pca=False)
kde_model.train_model(train_data)
samples = kde_model.generate_sample(N_SAMPLES)
fig = plot(train_data, samples)
plt.savefig('gaussian_kde.png', bbox_inches='tight')
plt.close(fig)
# TRAIN GAN #
gan_model = GAN(input_size=1, random_size=1)
gan_model.init_training()
# Create dataset
train_data_m = Dataset(train_data)
gan_model.train_model(train_data_m)
samples = gan_model.generate_sample(N_SAMPLES)
fig = plot(train_data, samples)
plt.savefig('gaussian_gan.png', bbox_inches='tight')
plt.close(fig)
ax.set_xticklabels([])
ax.set_yticklabels([])
ax.set_aspect('equal')
plt.imshow(sample.reshape(28, 28), cmap='Greys_r')
return fig
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
train_data = mnist.train.images[0:2000, :]
# TRAIN KDE #
kde_model = KDE()
kde_model.train_model(train_data)
samples = kde_model.generate_sample(16)
fig = plot(samples)
plt.savefig('kde_mnist.png', bbox_inches='tight')
plt.close(fig)
# TRAIN GAN #
gan_model = GAN()
gan_model.init_training()
gan_model.train_model(mnist)
samples = gan_model.generate_sample(16)
fig = plot(samples)
plt.savefig('gan_mnist.png', bbox_inches='tight')
plt.close(fig)
