270 lines
11 KiB
Python
270 lines
11 KiB
Python
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
import torch
|
|
from torch.optim import Optimizer
|
|
import math
|
|
|
|
|
|
def lr_policy(step, epoch, initial_lr, optimizer, steps_per_epoch, warmup_epochs,
|
|
hold_epochs, num_epochs=None, policy='linear', min_lr=1e-5,
|
|
exp_gamma=None):
|
|
"""
|
|
learning rate decay
|
|
Args:
|
|
initial_lr: base learning rate
|
|
step: current iteration number
|
|
N: total number of iterations over which learning rate is decayed
|
|
lr_steps: list of steps to apply exp_gamma
|
|
"""
|
|
warmup_steps = warmup_epochs * steps_per_epoch
|
|
hold_steps = hold_epochs * steps_per_epoch
|
|
|
|
if policy == 'legacy':
|
|
assert num_epochs is not None
|
|
tot_steps = num_epochs * steps_per_epoch
|
|
|
|
if step < warmup_steps:
|
|
a = (step + 1) / (warmup_steps + 1)
|
|
elif step < warmup_steps + hold_steps:
|
|
a = 1.0
|
|
else:
|
|
a = (((tot_steps - step)
|
|
/ (tot_steps - warmup_steps - hold_steps)) ** 2)
|
|
|
|
elif policy == 'exponential':
|
|
assert exp_gamma is not None
|
|
|
|
if step < warmup_steps:
|
|
a = (step + 1) / (warmup_steps + 1)
|
|
elif step < warmup_steps + hold_steps:
|
|
a = 1.0
|
|
else:
|
|
a = exp_gamma ** (epoch - warmup_epochs - hold_epochs)
|
|
|
|
else:
|
|
raise ValueError
|
|
|
|
new_lr = max(a * initial_lr, min_lr)
|
|
for param_group in optimizer.param_groups:
|
|
param_group['lr'] = new_lr
|
|
|
|
|
|
class AdamW(Optimizer):
|
|
"""Implements AdamW algorithm.
|
|
|
|
It has been proposed in `Adam: A Method for Stochastic Optimization`_.
|
|
|
|
Arguments:
|
|
params (iterable): iterable of parameters to optimize or dicts defining
|
|
parameter groups
|
|
lr (float, optional): learning rate (default: 1e-3)
|
|
betas (Tuple[float, float], optional): coefficients used for computing
|
|
running averages of gradient and its square (default: (0.9, 0.999))
|
|
eps (float, optional): term added to the denominator to improve
|
|
numerical stability (default: 1e-8)
|
|
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
|
|
amsgrad (boolean, optional): whether to use the AMSGrad variant of this
|
|
algorithm from the paper `On the Convergence of Adam and Beyond`_
|
|
|
|
Adam: A Method for Stochastic Optimization:
|
|
https://arxiv.org/abs/1412.6980
|
|
On the Convergence of Adam and Beyond:
|
|
https://openreview.net/forum?id=ryQu7f-RZ
|
|
"""
|
|
|
|
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8,
|
|
weight_decay=0, amsgrad=False):
|
|
if not 0.0 <= lr:
|
|
raise ValueError("Invalid learning rate: {}".format(lr))
|
|
if not 0.0 <= eps:
|
|
raise ValueError("Invalid epsilon value: {}".format(eps))
|
|
if not 0.0 <= betas[0] < 1.0:
|
|
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
|
|
if not 0.0 <= betas[1] < 1.0:
|
|
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
|
|
defaults = dict(lr=lr, betas=betas, eps=eps,
|
|
weight_decay=weight_decay, amsgrad=amsgrad)
|
|
super(AdamW, self).__init__(params, defaults)
|
|
|
|
def __setstate__(self, state):
|
|
super(AdamW, self).__setstate__(state)
|
|
for group in self.param_groups:
|
|
group.setdefault('amsgrad', False)
|
|
|
|
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:
|
|
for p in group['params']:
|
|
if p.grad is None:
|
|
continue
|
|
grad = p.grad.data
|
|
if grad.is_sparse:
|
|
raise RuntimeError('Adam does not support sparse gradients, please consider SparseAdam instead')
|
|
amsgrad = group['amsgrad']
|
|
|
|
state = self.state[p]
|
|
|
|
# State initialization
|
|
if len(state) == 0:
|
|
state['step'] = 0
|
|
# Exponential moving average of gradient values
|
|
state['exp_avg'] = torch.zeros_like(p.data)
|
|
# Exponential moving average of squared gradient values
|
|
state['exp_avg_sq'] = torch.zeros_like(p.data)
|
|
if amsgrad:
|
|
# Maintains max of all exp. moving avg. of sq. grad. values
|
|
state['max_exp_avg_sq'] = torch.zeros_like(p.data)
|
|
|
|
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
|
if amsgrad:
|
|
max_exp_avg_sq = state['max_exp_avg_sq']
|
|
beta1, beta2 = group['betas']
|
|
|
|
state['step'] += 1
|
|
# Decay the first and second moment running average coefficient
|
|
exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1)
|
|
exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
|
|
if amsgrad:
|
|
# Maintains the maximum of all 2nd moment running avg. till now
|
|
torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
|
|
# Use the max. for normalizing running avg. of gradient
|
|
denom = max_exp_avg_sq.sqrt().add_(group['eps'])
|
|
else:
|
|
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
|
|
|
bias_correction1 = 1 - beta1 ** state['step']
|
|
bias_correction2 = 1 - beta2 ** state['step']
|
|
step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
|
|
p.data.add_(torch.mul(p.data, group['weight_decay']).addcdiv_(1, exp_avg, denom), alpha=-step_size)
|
|
|
|
return loss
|
|
|
|
|
|
class Novograd(Optimizer):
|
|
"""
|
|
Implements Novograd algorithm.
|
|
|
|
Args:
|
|
params (iterable): iterable of parameters to optimize or dicts defining
|
|
parameter groups
|
|
lr (float, optional): learning rate (default: 1e-3)
|
|
betas (Tuple[float, float], optional): coefficients used for computing
|
|
running averages of gradient and its square (default: (0.95, 0))
|
|
eps (float, optional): term added to the denominator to improve
|
|
numerical stability (default: 1e-8)
|
|
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
|
|
grad_averaging: gradient averaging
|
|
amsgrad (boolean, optional): whether to use the AMSGrad variant of this
|
|
algorithm from the paper `On the Convergence of Adam and Beyond`_
|
|
(default: False)
|
|
"""
|
|
|
|
def __init__(self, params, lr=1e-3, betas=(0.95, 0), eps=1e-8,
|
|
weight_decay=0, grad_averaging=False, amsgrad=False):
|
|
if not 0.0 <= lr:
|
|
raise ValueError("Invalid learning rate: {}".format(lr))
|
|
if not 0.0 <= eps:
|
|
raise ValueError("Invalid epsilon value: {}".format(eps))
|
|
if not 0.0 <= betas[0] < 1.0:
|
|
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
|
|
if not 0.0 <= betas[1] < 1.0:
|
|
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
|
|
defaults = dict(lr=lr, betas=betas, eps=eps,
|
|
weight_decay=weight_decay,
|
|
grad_averaging=grad_averaging,
|
|
amsgrad=amsgrad)
|
|
|
|
super(Novograd, self).__init__(params, defaults)
|
|
|
|
def __setstate__(self, state):
|
|
super(Novograd, self).__setstate__(state)
|
|
for group in self.param_groups:
|
|
group.setdefault('amsgrad', False)
|
|
|
|
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:
|
|
for p in group['params']:
|
|
if p.grad is None:
|
|
continue
|
|
grad = p.grad.data
|
|
if grad.is_sparse:
|
|
raise RuntimeError('Sparse gradients are not supported.')
|
|
amsgrad = group['amsgrad']
|
|
|
|
state = self.state[p]
|
|
|
|
# State initialization
|
|
if len(state) == 0:
|
|
state['step'] = 0
|
|
# Exponential moving average of gradient values
|
|
state['exp_avg'] = torch.zeros_like(p.data)
|
|
# Exponential moving average of squared gradient values
|
|
state['exp_avg_sq'] = torch.zeros([]).to(state['exp_avg'].device)
|
|
if amsgrad:
|
|
# Maintains max of all exp. moving avg. of sq. grad. values
|
|
state['max_exp_avg_sq'] = torch.zeros([]).to(state['exp_avg'].device)
|
|
|
|
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
|
|
if amsgrad:
|
|
max_exp_avg_sq = state['max_exp_avg_sq']
|
|
beta1, beta2 = group['betas']
|
|
|
|
state['step'] += 1
|
|
|
|
norm = torch.sum(torch.pow(grad, 2))
|
|
|
|
if exp_avg_sq == 0:
|
|
exp_avg_sq.copy_(norm)
|
|
else:
|
|
exp_avg_sq.mul_(beta2).add_(norm, alpha=1 - beta2)
|
|
|
|
if amsgrad:
|
|
# Maintains the maximum of all 2nd moment running avg. till now
|
|
torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
|
|
# Use the max. for normalizing running avg. of gradient
|
|
denom = max_exp_avg_sq.sqrt().add_(group['eps'])
|
|
else:
|
|
denom = exp_avg_sq.sqrt().add_(group['eps'])
|
|
|
|
grad.div_(denom)
|
|
if group['weight_decay'] != 0:
|
|
grad.add_(p.data, alpha=group['weight_decay'])
|
|
if group['grad_averaging']:
|
|
grad.mul_(1 - beta1)
|
|
exp_avg.mul_(beta1).add_(grad)
|
|
|
|
p.data.add_(exp_avg, alpha=-group['lr'])
|
|
|
|
return loss
|