lars.py

## LARS SGD Optimizer as used in the paper:

""" Layer-wise adaptive rate scaling for SGD in PyTorch! """
import torch
from torch.optim.optimizer import Optimizer, required

class LARS(Optimizer):
    r"""Implements layer-wise adaptive rate scaling for SGD.
    Args:
        params (iterable): iterable of parameters to optimize or dicts defining
            parameter groups
        lr (float): base learning rate (\gamma_0)
        momentum (float, optional): momentum factor (default: 0) ("m")
        weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
            ("\beta")
        dampening (float, optional): dampening for momentum (default: 0)
        eta (float, optional): LARS coefficient
        nesterov (bool, optional): enables Nesterov momentum (default: False)
    Based on Algorithm 1 of the following paper by You, Gitman, and Ginsburg.
    Large Batch Training of Convolutional Networks:
        https://arxiv.org/abs/1708.03888
    Example:
        >>> optimizer = LARS(model.parameters(), lr=0.1, momentum=0.9,
        >>>                  weight_decay=1e-4, eta=1e-3)
        >>> optimizer.zero_grad()
        >>> loss_fn(model(input), target).backward()
        >>> optimizer.step()
    """

    def __init__(self,
                 params,
                 lr=required,
                 momentum=0,
                 dampening=0,
                 weight_decay=0,
                 eta=0.001,
                 nesterov=False):
        if lr is not required and lr < 0.0:
            raise ValueError("Invalid learning rate: {}".format(lr))
        if momentum < 0.0:
            raise ValueError("Invalid momentum value: {}".format(momentum))
        if weight_decay < 0.0:
            raise ValueError(
                "Invalid weight_decay value: {}".format(weight_decay))
        if eta < 0.0:
            raise ValueError("Invalid LARS coefficient value: {}".format(eta))

        defaults = dict(
            lr=lr, momentum=momentum, dampening=dampening,
            weight_decay=weight_decay, nesterov=nesterov, eta=eta)
        if nesterov and (momentum <= 0 or dampening != 0):
            raise ValueError("Nesterov momentum requires a momentum and zero dampening")

        super(LARS, self).__init__(params, defaults)

    def __setstate__(self, state):
        super(LARS, self).__setstate__(state)
        for group in self.param_groups:
            group.setdefault('nesterov', False)

    @torch.no_grad()
    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:
            with torch.enable_grad():
                loss = closure()

        for group in self.param_groups:
            weight_decay = group['weight_decay']
            momentum = group['momentum']
            dampening = group['dampening']
            eta = group['eta']
            nesterov = group['nesterov']
            lr = group['lr']
            lars_exclude = group.get('lars_exclude', False)

            for p in group['params']:
                if p.grad is None:
                    continue

                d_p = p.grad

                if lars_exclude:
                    local_lr = 1.
                else:
                    weight_norm = torch.norm(p).item()
                    grad_norm = torch.norm(d_p).item()
                    # Compute local learning rate for this layer
                    local_lr = eta * weight_norm / \
                        (grad_norm + weight_decay * weight_norm)

                actual_lr = local_lr * lr
                d_p = d_p.add(p, alpha=weight_decay).mul(actual_lr)
                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_(d_p, alpha=1 - dampening)
                    if nesterov:
                        d_p = d_p.add(buf, alpha=momentum)
                    else:
                        d_p = buf
                p.add_(-d_p)

        return loss