def test_warmup_multistep(self):
p = nn.Parameter(torch.zeros(0))
opt = torch.optim.SGD([p], lr=5)
multiplier = WarmupParamScheduler(
MultiStepParamScheduler(
[1, 0.1, 0.01, 0.001],
milestones=[10, 15, 20],
num_updates=30,
),
0.001,
5 / 30,
)
sched = LRMultiplier(opt, multiplier, 30)
# This is an equivalent of:
# sched = WarmupMultiStepLR(
# opt, milestones=[10, 15, 20], gamma=0.1, warmup_factor=0.001, warmup_iters=5)
p.sum().backward()
opt.step()
lrs = [0.005]
for _ in range(30):
sched.step()
lrs.append(opt.param_groups[0]["lr"])
self.assertTrue(
np.allclose(lrs[:5], [0.005, 1.004, 2.003, 3.002, 4.001]))
self.assertTrue(np.allclose(lrs[5:10], 5.0))
self.assertTrue(np.allclose(lrs[10:15], 0.5))
self.assertTrue(np.allclose(lrs[15:20], 0.05))
self.assertTrue(np.allclose(lrs[20:], 0.005))
def test_warmup_cosine(self):
p = nn.Parameter(torch.zeros(0))
opt = torch.optim.SGD([p], lr=5)
multiplier = CompositeParamScheduler(
[
LinearParamScheduler(0.001, 1), # warmup
CosineParamScheduler(1, 0),
],
interval_scaling=["rescaled", "fixed"],
lengths=[5 / 30, 25 / 30],
)
sched = LRMultiplier(opt, multiplier, 30)
p.sum().backward()
opt.step()
self.assertEqual(opt.param_groups[0]["lr"], 0.005)
lrs = [0.005]
for _ in range(30):
sched.step()
lrs.append(opt.param_groups[0]["lr"])
for idx, lr in enumerate(lrs):
expected_cosine = 2.5 * (1.0 + math.cos(math.pi * idx / 30))
if idx >= 5:
self.assertAlmostEqual(lr, expected_cosine)
else:
self.assertNotAlmostEqual(lr, expected_cosine)
def before_train(self):
self._optimizer = self._optimizer or self.trainer.optimizer
self._scheduler = self._scheduler or self.trainer.scheduler
if isinstance(self._scheduler, ParamScheduler):
self._scheduler = LRMultiplier(
self._optimizer,
self._scheduler,
self.trainer.max_iter,
last_iter=self.trainer.iter - 1,
)
# NOTE: some heuristics on what LR to summarize
# summarize the param group with most parameters
largest_group = max(len(g["params"]) for g in self._optimizer.param_groups)
if largest_group == 1:
# If all groups have one parameter,
# then find the most common initial LR, and use it for summary
lr_count = Counter([g["lr"] for g in self._optimizer.param_groups])
lr = lr_count.most_common()[0][0]
for i, g in enumerate(self._optimizer.param_groups):
if g["lr"] == lr:
self._best_param_group_id = i
break
else:
for i, g in enumerate(self._optimizer.param_groups):
if len(g["params"]) == largest_group:
self._best_param_group_id = i
break
class LRScheduler(HookBase):
"""
A hook which executes a torch builtin LR scheduler and summarizes the LR.
It is executed after every iteration.
"""
def __init__(self, optimizer=None, scheduler=None):
"""
Args:
optimizer (torch.optim.Optimizer):
scheduler (torch.optim.LRScheduler or fvcore.common.param_scheduler.ParamScheduler):
if a :class:`ParamScheduler` object, it defines the multiplier over the base LR
in the optimizer.
If any argument is not given, will try to obtain it from the trainer.
"""
self._optimizer = optimizer
self._scheduler = scheduler
def before_train(self):
self._optimizer = self._optimizer or self.trainer.optimizer
self._scheduler = self._scheduler or self.trainer.scheduler
if isinstance(self._scheduler, ParamScheduler):
self._scheduler = LRMultiplier(
self._optimizer,
self._scheduler,
self.trainer.max_iter,
last_iter=self.trainer.iter - 1,
)
# NOTE: some heuristics on what LR to summarize
# summarize the param group with most parameters
largest_group = max(len(g["params"]) for g in self._optimizer.param_groups)
if largest_group == 1:
# If all groups have one parameter,
# then find the most common initial LR, and use it for summary
lr_count = Counter([g["lr"] for g in self._optimizer.param_groups])
lr = lr_count.most_common()[0][0]
for i, g in enumerate(self._optimizer.param_groups):
if g["lr"] == lr:
self._best_param_group_id = i
break
else:
for i, g in enumerate(self._optimizer.param_groups):
if len(g["params"]) == largest_group:
self._best_param_group_id = i
break
def after_step(self):
lr = self._optimizer.param_groups[self._best_param_group_id]["lr"]
self.trainer.storage.put_scalar("lr", lr, smoothing_hint=False)
self._scheduler.step()
def before_train(self):
self._optimizer = self._optimizer or self.trainer.optimizer
if isinstance(self.scheduler, ParamScheduler):
self._scheduler = LRMultiplier(
self._optimizer,
self.scheduler,
self.trainer.max_iter,
last_iter=self.trainer.iter - 1,
)
self._best_param_group_id = LRScheduler.get_best_param_group_id(self._optimizer)
def test_warmup_cosine(self):
p = nn.Parameter(torch.zeros(0))
opt = torch.optim.SGD([p], lr=5)
multiplier = WarmupParamScheduler(
CosineParamScheduler(1, 0),
0.001,
5 / 30,
)
sched = LRMultiplier(opt, multiplier, 30)
p.sum().backward()
opt.step()
self.assertEqual(opt.param_groups[0]["lr"], 0.005)
lrs = [0.005]
for _ in range(30):
sched.step()
lrs.append(opt.param_groups[0]["lr"])
for idx, lr in enumerate(lrs):
expected_cosine = 2.5 * (1.0 + math.cos(math.pi * idx / 30))
if idx >= 5:
self.assertAlmostEqual(lr, expected_cosine)
else:
self.assertNotAlmostEqual(lr, expected_cosine)