def get_reconstruction_loss(self, x):
hx = ilr(self.imputer(x), self.Psi)
z_mean = self.encoder(hx)
eta = self.decoder(z_mean)
logp = self.Psi.t() @ eta.t()
mult_loss = Multinomial(logits=logp.t()).log_prob(x).mean()
return -mult_loss
def encode(self, x):
#B = B.sum(axis=0) + 1
#B = B.unsqueeze(0)
#batch_effects = (self.Psi @ B.t()).t()
hx = ilr(self.imputer(x), self.Psi)
#hx -= batch_effects # Subtract out batch effects
z = self.encoder(hx)
return z
def get_reconstruction_loss(self, x, B):
hx = ilr(self.imputer(x), self.Psi)
batch_effects = (self.Psi @ B.t()).t()
hx -= batch_effects # Subtract out batch effects
z_mean = self.encoder(hx)
eta = self.decoder(z_mean)
eta += batch_effects # Add batch effects back in
recon_loss = -self.recon_model_loglik(x, eta)
return recon_loss
def get_reconstruction_loss(self, x, B):
hx = ilr(self.imputer(x), self.Psi)
batch_effects = (self.Psi @ B.t()).t()
hx -= batch_effects # Subtract out batch effects
z_mean = self.encoder(hx)
eta = self.decoder(z_mean)
eta += batch_effects # Add batch effects back in
logp = self.Psi.t() @ eta.t()
mult_loss = Multinomial(logits=logp.t()).log_prob(x).mean()
return -mult_loss
def get_reconstruction_loss(self, x):
x_ = ilr(self.imputer(x), self.Psi)
if self.use_analytic_elbo:
return -self.analytic_exp_recon_loss(x_)
else:
z_mean = self.encoder(x_)
eps = torch.normal(torch.zeros_like(z_mean), 1.0)
z_sample = z_mean + eps * torch.exp(0.5 * self.variational_logvars)
x_out = self.decoder(z_sample)
recon_loss = -self.recon_model_loglik(x, x_out)
return recon_loss
def forward(self, x, B):
hx = ilr(self.imputer(x), self.Psi)
batch_effects = (self.Psi @ B.t()).t()
hx -= batch_effects # Subtract out batch effects
z_mean = self.encoder(hx)
eps = torch.normal(torch.zeros_like(z_mean), 1.0)
z_sample = z_mean + eps * torch.exp(0.5 * self.variational_logvars)
x_out = self.decoder(z_sample)
x_out += batch_effects # Add batch effects back in
kl_div = (
-self.gaussian_kl(z_mean, self.variational_logvars)).mean(0).sum()
recon_loss = (-self.recon_model_loglik(x, x_out)).mean(0).sum()
loss = kl_div + recon_loss
return loss
def forward(self, x):
hx = ilr(self.imputer(x), self.Psi)
z_mean = self.encoder(hx)
mu = self.decoder(z_mean)
W = self.decoder.weight
# penalties
D = torch.exp(self.variational_logvars)
var = torch.exp(self.log_sigma_sq)
qdist = MultivariateNormalFactorIdentity(mu, var, D, W)
logp = self.Psi.t() @ self.eta.t()
prior_loss = Normal(self.zm, self.zI).log_prob(z_mean).mean()
logit_loss = qdist.log_prob(self.eta).mean()
mult_loss = Multinomial(logits=logp.t()).log_prob(x).mean()
loglike = mult_loss + logit_loss + prior_loss
return -loglike
def forward(self, x):
x_ = ilr(self.imputer(x), self.Psi)
z_mean = self.encoder(x_)
if not self.use_analytic_elbo:
eps = torch.normal(torch.zeros_like(z_mean), 1.0)
z_sample = z_mean + eps * torch.exp(0.5 * self.variational_logvars)
x_out = self.decoder(z_sample)
kl_div = (-self.gaussian_kl(
z_mean, self.variational_logvars)).mean(0).sum()
recon_loss = (-self.recon_model_loglik(x, x_out)).mean(0).sum()
loss = kl_div + recon_loss
else:
loss = self.analytic_elbo(x_, z_mean)
return loss
def encode(self, x):
hx = ilr(self.imputer(x), self.Psi)
z = self.encoder(hx)
return z
def reset(self, x):
hx = ilr(self.imputer(x), self.Psi)
self.eta.data = hx.data