def __init__(self, binary_features, continuous_features, z_dim, hidden_dim,
hidden_layers, optimizer, activation, cuda):
pyro.clear_param_store()
vae = VAE(binary_features, continuous_features, z_dim, hidden_dim,
hidden_layers, activation, cuda)
vae = vae.double()
self.vae = vae
self.svi = SVI(vae.model, vae.guide, optimizer, loss=Trace_ELBO())
self.cuda = cuda
self.train_stats = Statistics()
self.test_stats = Statistics()
mean_asian = np.mean(asian_points, axis=0)
mean_male = np.mean(male_points, axis=0)
mean_female = np.mean(female_points, axis=0)
var_white = np.cov(white_points, rowvar=False)
var_black = np.cov(black_points, rowvar=False)
var_asian = np.cov(asian_points, rowvar=False)
var_male = np.cov(male_points, rowvar=False)
var_female = np.cov(female_points, rowvar=False)
full_mean = np.mean(mu_points, axis=0)
full_var = np.cov(mu_points, rowvar=False)
# <codecell>
# TODO: redo VAE with permute instead of reshape
model = model.double()
with torch.no_grad():
white_im = model.decode(torch.from_numpy(mean_white)).numpy()
black_im = model.decode(torch.from_numpy(mean_black)).numpy()
asian_im = model.decode(torch.from_numpy(mean_asian)).numpy()
male_im = model.decode(torch.from_numpy(mean_male)).numpy()
female_im = model.decode(torch.from_numpy(mean_female)).numpy()
full_im = model.decode(torch.from_numpy(full_mean)).numpy()
white_im = np.squeeze(white_im).transpose(1, 2, 0)
plt.title('Average across White faces')
plt.imshow(white_im)
plt.savefig(save_path / 'white_mean.png')
black_im = np.squeeze(black_im).transpose(1, 2, 0)
plt.title('Average across Black faces')
