45a03ac258234ba49c0a43d46ae022493190591e,torch_geometric/nn/models/autoencoder.py,ARGA,discriminator_loss,#ARGA#Any#Any#,220

Before Change

    def discriminator_loss(self, d_real, d_fake):
        dc_real_loss = torch.nn.BCELoss(reduction="mean")(
            d_real, torch.ones(d_real.size()))
        dc_fake_loss = torch.nn.BCELoss(reduction="mean")(
            d_fake, torch.zeros(d_fake.size()))
        dc_gen_loss = torch.nn.BCELoss(reduction="mean")(
            d_fake, torch.ones(d_fake.size()))
        return dc_real_loss + dc_fake_loss + dc_gen_loss

    def loss(self, d_real, d_fake, *args):
        args = list(args)
        args[0] = self.decoder(args[0])

After Change

        return real, fake

    def discriminator_loss(self, real, fake):
        real_loss = -torch.log(real).mean()
        fake_loss = -torch.log((1 - fake).clamp(min=1e-8)).mean()
        return real_loss + fake_loss

    def loss(self, z, pos_edge_index, neg_adj_mask):
        recon_loss = self.reconstruction_loss(z, pos_edge_index, neg_adj_mask)
        d_loss = self.discriminator_loss(*self.discriminate(z))

Instances