def __init__()

in domainbed/algorithms.py [0:0]

• 28 lines of code
• 1 McCabe index (conditional complexity)

    def __init__(self, input_shape, num_classes, num_domains,
                 hparams, conditional, class_balance):

        super(AbstractDANN, self).__init__(input_shape, num_classes, num_domains,
                                  hparams)

        self.register_buffer('update_count', torch.tensor([0]))
        self.conditional = conditional
        self.class_balance = class_balance

        # Algorithms
        self.featurizer = networks.Featurizer(input_shape, self.hparams)
        self.classifier = networks.Classifier(
            self.featurizer.n_outputs,
            num_classes,
            self.hparams['nonlinear_classifier'])
        self.discriminator = networks.MLP(self.featurizer.n_outputs,
            num_domains, self.hparams)
        self.class_embeddings = nn.Embedding(num_classes,
            self.featurizer.n_outputs)

        # Optimizers
        self.disc_opt = torch.optim.Adam(
            (list(self.discriminator.parameters()) +
                list(self.class_embeddings.parameters())),
            lr=self.hparams["lr_d"],
            weight_decay=self.hparams['weight_decay_d'],
            betas=(self.hparams['beta1'], 0.9))

        self.gen_opt = torch.optim.Adam(
            (list(self.featurizer.parameters()) +
                list(self.classifier.parameters())),
            lr=self.hparams["lr_g"],
            weight_decay=self.hparams['weight_decay_g'],
            betas=(self.hparams['beta1'], 0.9))