SGD Nesterov for Optim

Any idea why nesterov is not available under optim? Seems to be available under legacy here: https://github.com/pytorch/pytorch/blob/master/torch/legacy/optim/sgd.py

Does it mean if I’d like to have nesterov, I’ve to modify PyTorch optim.SGD?

Found a suggestion on Github by ajbrock to change it to:

from .optimizer import Optimizer, required


class SGD(Optimizer):
    """Implements stochastic gradient descent (optionally with momentum).
    Args:
        params (iterable): iterable of parameters to optimize or dicts defining
            parameter groups
        lr (float): learning rate
        momentum (float, optional): momentum factor (default: 0)
        weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
        dampening (float, optional): dampening for momentum (default: 0)
        nesterov(bool, optional): enables Nesterov momentum (default: False)
        
    Example:
        >>> optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.9)
        >>> optimizer.zero_grad()
        >>> loss_fn(model(input), target).backward()
        >>> optimizer.step()
    """

    def __init__(self, params, lr=required, momentum=0, dampening=0,
                 weight_decay=0, nesterov=False):
        defaults = dict(lr=lr, momentum=momentum, dampening=dampening,
                        weight_decay=weight_decay, nesterov=nesterov)
        if nesterov and (momentum <= 0 and dampening != 0):
            raise ValueError("Nesterov momentum requires a momentum and zero dampening")
        super(SGD, self).__init__(params, defaults)

    def step(self, 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()

        for group in self.param_groups:
            weight_decay = group['weight_decay']
            momentum = group['momentum']
            dampening = group['dampening']
            nesterov = group['nesterov']

            for p in group['params']:
                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:
                        param_state['momentum_buffer'] = d_p.clone()
                    else:
                        buf = param_state['momentum_buffer']
                        buf.mul_(momentum).add_(1 - dampening, d_p)
                        if nesterov:
                            d_p.add_(momentum, buf)
                        else:
                            d_p = buf

                p.data.add_(-group['lr'], d_p)

        return loss

Yes, it’s going to be merged into master soon.

3 Likes

Why is the Nesterov method requires Momentum?
Is it implemented in Nesterov way or in the simpler form of FISTA?

Accelerated gradient descent is not a momentum method, but it has been shown that it is closely related and the update rule can be rewritten as a momentum-like update rule.

I think that in deep learning literature, the method has been introduced by Ilya Sutskever, and since then, the implementations are closely based on the original paper: http://www.cs.toronto.edu/~fritz/absps/momentum.pdf

1 Like