import torch
from torch.optim.optimizer import Optimizer, required
from torch import optim, nn
from nfnets.utils import unitwise_norm
[docs]class SGD_AGC(Optimizer):
r"""Implements stochastic gradient descent (optionally with momentum).
Nesterov momentum is based on the formula from
`On the importance of initialization and momentum in deep learning`__
AGC from NFNets: https://arxiv.org/abs/2102.06171.pdf.
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)
dampening (float, optional): dampening for momentum (default: 0.01)
clipping (float, optional): clipping value (default: 1e-3)
eps (float, optional): eps (default: 1e-3)
Example:
>>> optimizer = torch.optim.SGD_AGC(model.parameters(), lr=0.1, momentum=0.9)
>>> optimizer.zero_grad()
>>> loss_fn(model(input), target).backward()
>>> optimizer.step()
.. note::
The implementation has been adapted from the PyTorch framework and the official
NF-Nets paper.
The implementation of SGD with Momentum/Nesterov subtly differs from
Sutskever et. al. and implementations in some other frameworks.
Considering the specific case of Momentum, the update can be written as
.. math::
\begin{aligned}
v_{t+1} & = \mu * v_{t} + g_{t+1}, \\
p_{t+1} & = p_{t} - \text{lr} * v_{t+1},
\end{aligned}
where :math:`p`, :math:`g`, :math:`v` and :math:`\mu` denote the
parameters, gradient, velocity, and momentum respectively.
This is in contrast to Sutskever et. al. and
other frameworks which employ an update of the form
.. math::
\begin{aligned}
v_{t+1} & = \mu * v_{t} + \text{lr} * g_{t+1}, \\
p_{t+1} & = p_{t} - v_{t+1}.
\end{aligned}
The Nesterov version is analogously modified.
"""
def __init__(self, params, lr=required, momentum=0, dampening=0,
weight_decay=0, nesterov=False, clipping=1e-2, eps=1e-3):
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 clipping < 0.0:
raise ValueError("Invalid clipping value: {}".format(clipping))
if eps < 0.0:
raise ValueError("Invalid eps value: {}".format(eps))
defaults = dict(lr=lr, momentum=momentum, dampening=dampening,
weight_decay=weight_decay, nesterov=nesterov, clipping=clipping, eps=eps)
if nesterov and (momentum <= 0 or dampening != 0):
raise ValueError(
"Nesterov momentum requires a momentum and zero dampening")
super(SGD_AGC, self).__init__(params, defaults)
def __setstate__(self, state):
super(SGD_AGC, self).__setstate__(state)
for group in self.param_groups:
group.setdefault('nesterov', False)
[docs] @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:
for p in group['params']:
if p.grad is None:
continue
param_norm = torch.max(unitwise_norm(
p.detach()), torch.tensor(group['eps']).to(p.device))
grad_norm = unitwise_norm(p.grad.detach())
max_norm = param_norm * group['clipping']
trigger = grad_norm > max_norm
clipped_grad = p.grad * \
(max_norm / torch.max(grad_norm,
torch.tensor(1e-6).to(grad_norm.device)))
p.grad.detach().copy_(torch.where(trigger, clipped_grad, p.grad))
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']:
if p.grad is None:
continue
d_p = p.grad
if weight_decay != 0:
d_p = d_p.add(p, alpha=weight_decay)
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, alpha=-group['lr'])
return loss