Method step

optimizers/sgd.py:261–305

Performs a single optimization step. Arguments: closure (callable, optional): A closure that reevaluates the model and returns the loss.

(self, layers_index_todo, lr_vector, closure=None)

            group.setdefault('nesterov', False)

    def step(self, layers_index_todo, lr_vector, closure=None):
        """Performs a single optimization step.

        Arguments:
            closure (callable, optional): A closure that reevaluates the model
                and returns the loss.
        """
        loss = None
        if closure is not None:
            loss = closure()

        iteration_group = 0
        for group in self.param_groups:
            iteration_group += 1
            weight_decay = group['weight_decay']
            momentum = group['momentum']
            dampening = group['dampening']
            nesterov = group['nesterov']

            iteration_p = 0
            for p in group['params']:
                if p.grad is None or ~layers_index_todo[iteration_p]:
                    iteration_p += 1
                    continue
                d_p = p.grad.data
                if weight_decay != 0:
                    d_p.add_(weight_decay, p.data)
                if momentum != 0:
                    param_state = self.state[p]
                    if 'momentum_buffer' not in param_state:
                        buf = param_state['momentum_buffer'] = torch.clone(
                            d_p).detach()
                    else:
                        buf = param_state['momentum_buffer']
                        buf.mul_(momentum).add_(1 - dampening, d_p)
                    if nesterov:
                        d_p = d_p.add(momentum, buf)
                    else:
                        d_p = buf

                # p.data.add_(-group['lr'], d_p)
                p.data.add_(d_p, alpha=-lr_vector[iteration_p])
                iteration_p += 1

        return loss