lightning/tests/tests_pytorch/tuner/test_lr_finder.py

540 lines
18 KiB
Python

# Copyright The Lightning AI team.
#
# 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 logging
import math
import os
from copy import deepcopy
from typing import Any
from unittest import mock
import pytest
import torch
from lightning_utilities.test.warning import no_warning_call
from lightning.pytorch import seed_everything, Trainer
from lightning.pytorch.callbacks.lr_finder import LearningRateFinder
from lightning.pytorch.demos.boring_classes import BoringModel
from lightning.pytorch.tuner.lr_finder import _LRFinder
from lightning.pytorch.tuner.tuning import Tuner
from lightning.pytorch.utilities.exceptions import MisconfigurationException
from lightning.pytorch.utilities.types import STEP_OUTPUT
from tests_pytorch.helpers.datamodules import ClassifDataModule
from tests_pytorch.helpers.runif import RunIf
from tests_pytorch.helpers.simple_models import ClassificationModel
from tests_pytorch.helpers.utils import getattr_recursive
def test_error_with_multiple_optimizers(tmpdir):
"""Check that error is thrown when more than 1 optimizer is passed."""
class CustomBoringModel(BoringModel):
def __init__(self, lr):
super().__init__()
self.save_hyperparameters()
self.automatic_optimization = False
def configure_optimizers(self):
optimizer1 = torch.optim.SGD(self.parameters(), lr=self.hparams.lr)
optimizer2 = torch.optim.Adam(self.parameters(), lr=self.hparams.lr)
return [optimizer1, optimizer2]
model = CustomBoringModel(lr=1e-2)
trainer = Trainer(default_root_dir=tmpdir, max_epochs=1)
tuner = Tuner(trainer)
with pytest.raises(MisconfigurationException, match="only works with single optimizer"):
tuner.lr_find(model)
def test_model_reset_correctly(tmpdir):
"""Check that model weights are correctly reset after _lr_find()"""
model = BoringModel()
model.lr = 0.1
# logger file to get meta
trainer = Trainer(default_root_dir=tmpdir, max_epochs=1)
tuner = Tuner(trainer)
before_state_dict = deepcopy(model.state_dict())
tuner.lr_find(model, num_training=5)
after_state_dict = model.state_dict()
for key in before_state_dict:
assert torch.all(
torch.eq(before_state_dict[key], after_state_dict[key])
), "Model was not reset correctly after learning rate finder"
assert not any(f for f in os.listdir(tmpdir) if f.startswith(".lr_find"))
def test_trainer_reset_correctly(tmpdir):
"""Check that all trainer parameters are reset correctly after lr_find()"""
model = BoringModel()
model.lr = 0.1
# logger file to get meta
trainer = Trainer(default_root_dir=tmpdir, max_epochs=1)
tuner = Tuner(trainer)
changed_attributes = [
"accumulate_grad_batches",
"callbacks",
"checkpoint_callback",
"current_epoch",
"loggers",
"global_step",
"max_steps",
"fit_loop.max_steps",
"strategy.setup_optimizers",
"should_stop",
]
expected = {ca: getattr_recursive(trainer, ca) for ca in changed_attributes}
with no_warning_call(UserWarning, match="Please add the following callbacks"):
tuner.lr_find(model, num_training=5)
actual = {ca: getattr_recursive(trainer, ca) for ca in changed_attributes}
assert actual == expected
assert model.trainer == trainer
@pytest.mark.parametrize("use_hparams", [False, True])
def test_tuner_lr_find(tmpdir, use_hparams):
"""Test that lr_find updates the learning rate attribute."""
seed_everything(1)
class CustomBoringModel(BoringModel):
def __init__(self, lr):
super().__init__()
self.save_hyperparameters()
self.lr = lr
def configure_optimizers(self):
return torch.optim.SGD(self.parameters(), lr=self.hparams.lr if use_hparams else self.lr)
before_lr = 1e-2
model = CustomBoringModel(lr=before_lr)
trainer = Trainer(default_root_dir=tmpdir, max_epochs=2)
tuner = Tuner(trainer)
tuner.lr_find(model, update_attr=True)
after_lr = model.hparams.lr if use_hparams else model.lr
assert after_lr is not None
assert before_lr != after_lr, "Learning rate was not altered after running learning rate finder"
@pytest.mark.parametrize("use_hparams", [False, True])
def test_trainer_arg_str(tmpdir, use_hparams):
"""Test that setting trainer arg to string works."""
seed_everything(1)
class CustomBoringModel(BoringModel):
def __init__(self, my_fancy_lr):
super().__init__()
self.save_hyperparameters()
self.my_fancy_lr = my_fancy_lr
def configure_optimizers(self):
return torch.optim.SGD(self.parameters(), lr=self.hparams.my_fancy_lr if use_hparams else self.my_fancy_lr)
before_lr = 1e-2
model = CustomBoringModel(my_fancy_lr=before_lr)
trainer = Trainer(default_root_dir=tmpdir, max_epochs=2)
tuner = Tuner(trainer)
tuner.lr_find(model, update_attr=True, attr_name="my_fancy_lr")
after_lr = model.hparams.my_fancy_lr if use_hparams else model.my_fancy_lr
assert after_lr is not None
assert before_lr != after_lr, "Learning rate was not altered after running learning rate finder"
@pytest.mark.parametrize("opt", ["Adam", "Adagrad"])
def test_call_to_trainer_method(tmpdir, opt):
"""Test that directly calling the trainer method works."""
seed_everything(1)
class CustomBoringModel(BoringModel):
def __init__(self, lr):
super().__init__()
self.save_hyperparameters()
def configure_optimizers(self):
return (
torch.optim.Adagrad(self.parameters(), lr=self.hparams.lr)
if opt == "Adagrad"
else torch.optim.Adam(self.parameters(), lr=self.hparams.lr)
)
before_lr = 1e-2
model = CustomBoringModel(1e-2)
trainer = Trainer(default_root_dir=tmpdir, max_epochs=2)
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model, mode="linear")
after_lr = lr_finder.suggestion()
assert after_lr is not None
model.hparams.lr = after_lr
tuner.lr_find(model, update_attr=True)
assert after_lr is not None
assert before_lr != after_lr, "Learning rate was not altered after running learning rate finder"
@RunIf(sklearn=True)
@mock.patch.dict(os.environ, os.environ.copy(), clear=True)
def test_datamodule_parameter(tmpdir):
"""Test that the datamodule parameter works."""
seed_everything(1)
dm = ClassifDataModule()
model = ClassificationModel(lr=1e-3)
before_lr = model.lr
# logger file to get meta
trainer = Trainer(default_root_dir=tmpdir, max_epochs=2)
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model, datamodule=dm)
after_lr = lr_finder.suggestion()
model.lr = after_lr
assert after_lr is not None
assert before_lr != after_lr, "Learning rate was not altered after running learning rate finder"
def test_accumulation_and_early_stopping(tmpdir):
"""Test that early stopping of learning rate finder works, and that accumulation also works for this feature."""
seed_everything(1)
class TestModel(BoringModel):
def __init__(self):
super().__init__()
self.lr = 1e-3
model = TestModel()
trainer = Trainer(default_root_dir=tmpdir, accumulate_grad_batches=2)
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model, early_stop_threshold=None)
assert lr_finder.suggestion() != 1e-3
assert len(lr_finder.results["lr"]) == len(lr_finder.results["loss"]) == 100
assert lr_finder._total_batch_idx == 199
def test_suggestion_parameters_work(tmpdir):
"""Test that default skipping does not alter results in basic case."""
seed_everything(1)
class CustomBoringModel(BoringModel):
def __init__(self, lr):
super().__init__()
self.lr = lr
def configure_optimizers(self):
return torch.optim.SGD(self.parameters(), lr=self.lr)
# logger file to get meta
model = CustomBoringModel(lr=1e-2)
trainer = Trainer(default_root_dir=tmpdir, max_epochs=3)
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model)
lr1 = lr_finder.suggestion(skip_begin=10) # default
lr2 = lr_finder.suggestion(skip_begin=70) # way too high, should have an impact
assert lr1 is not None
assert lr2 is not None
assert lr1 != lr2, "Skipping parameter did not influence learning rate"
def test_suggestion_with_non_finite_values(tmpdir):
"""Test that non-finite values does not alter results."""
seed_everything(1)
class CustomBoringModel(BoringModel):
def __init__(self, lr):
super().__init__()
self.lr = lr
def configure_optimizers(self):
return torch.optim.SGD(self.parameters(), lr=self.lr)
model = CustomBoringModel(lr=1e-2)
trainer = Trainer(default_root_dir=tmpdir, max_epochs=3)
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model)
before_lr = lr_finder.suggestion()
lr_finder.results["loss"][-1] = float("nan")
after_lr = lr_finder.suggestion()
assert before_lr is not None
assert after_lr is not None
assert before_lr == after_lr, "Learning rate was altered because of non-finite loss values"
def test_lr_finder_fails_fast_on_bad_config(tmpdir):
"""Test that tune fails if the model does not have a lr BEFORE running lr find."""
trainer = Trainer(default_root_dir=tmpdir, max_steps=2)
tuner = Tuner(trainer)
with pytest.raises(AttributeError, match="should have one of these fields"):
tuner.lr_find(BoringModel(), update_attr=True)
def test_lr_candidates_between_min_and_max(tmpdir):
"""Test that learning rate candidates are between min_lr and max_lr."""
seed_everything(1)
class TestModel(BoringModel):
def __init__(self, learning_rate=0.1):
super().__init__()
self.save_hyperparameters()
model = TestModel()
trainer = Trainer(default_root_dir=tmpdir)
lr_min = 1e-8
lr_max = 1.0
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model, max_lr=lr_min, min_lr=lr_max, num_training=3)
lr_candidates = lr_finder.results["lr"]
assert all(lr_min <= lr <= lr_max for lr in lr_candidates)
def test_lr_finder_ends_before_num_training(tmpdir):
"""Tests learning rate finder ends before `num_training` steps."""
class TestModel(BoringModel):
def __init__(self, learning_rate=0.1):
super().__init__()
self.save_hyperparameters()
def on_before_optimizer_step(self, optimizer):
assert self.global_step < num_training
model = TestModel()
trainer = Trainer(default_root_dir=tmpdir)
tuner = Tuner(trainer)
num_training = 3
tuner.lr_find(model=model, num_training=num_training)
def test_multiple_lr_find_calls_gives_same_results(tmpdir):
"""Tests that lr_finder gives same results if called multiple times."""
seed_everything(1)
model = BoringModel()
model.lr = 0.1
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=2,
limit_train_batches=10,
limit_val_batches=2,
enable_progress_bar=False,
enable_model_summary=False,
enable_checkpointing=False,
)
tuner = Tuner(trainer)
all_res = [tuner.lr_find(model).results for _ in range(3)]
assert all(
all_res[0][k] == curr_lr_finder[k] and len(curr_lr_finder[k]) > 10
for curr_lr_finder in all_res[1:]
for k in all_res[0]
)
@pytest.mark.parametrize(
("skip_begin", "skip_end", "losses", "expected_error"),
[
(0, 0, [], True),
(10, 1, [], True),
(0, 2, [0, 1, 2], True),
(0, 1, [0, 1, 2], False),
(1, 1, [0, 1, 2], True),
(1, 1, [0, 1, 2, 3], False),
(0, 1, [float("nan"), float("nan"), 0, float("inf"), 1, 2, 3, float("inf"), 2, float("nan"), 1], False),
(4, 1, [float("nan"), float("nan"), 0, float("inf"), 1, 2, 3, float("inf"), 2, float("nan"), 1], False),
],
)
def test_suggestion_not_enough_finite_points(losses, skip_begin, skip_end, expected_error, caplog):
"""Tests the error handling when not enough finite points are available to make a suggestion."""
caplog.clear()
lr_finder = _LRFinder(
mode="exponential",
lr_min=1e-8,
lr_max=1,
num_training=100,
)
lrs = list(torch.arange(len(losses)))
lr_finder.results = {
"lr": lrs,
"loss": losses,
}
with caplog.at_level(logging.ERROR, logger="root.tuner.lr_finder"):
lr = lr_finder.suggestion(skip_begin=skip_begin, skip_end=skip_end)
if expected_error:
assert lr is None
assert "Failed to compute suggestion for learning rate" in caplog.text
else:
assert lr is not None
def test_lr_attribute_when_suggestion_invalid(tmpdir):
"""Tests learning rate finder ends before `num_training` steps."""
class TestModel(BoringModel):
def __init__(self):
super().__init__()
self.learning_rate = 0.123
model = TestModel()
trainer = Trainer(default_root_dir=tmpdir)
tuner = Tuner(trainer)
lr_finder = tuner.lr_find(model=model, update_attr=True, num_training=1) # force insufficient data points
assert lr_finder.suggestion() is None
assert model.learning_rate == 0.123 # must remain unchanged because suggestion is not possible
def test_lr_finder_callback_restarting(tmpdir):
"""Test that `LearningRateFinder` does not set restarting=True when loading checkpoint."""
num_lr_steps = 100
class MyBoringModel(BoringModel):
def __init__(self):
super().__init__()
self.learning_rate = 0.123
def on_train_batch_start(self, batch, batch_idx):
if getattr(self, "_expected_max_steps", None) is not None:
assert self.trainer.fit_loop.max_steps == self._expected_max_steps
def configure_optimizers(self):
return torch.optim.SGD(self.parameters(), lr=self.learning_rate)
class CustomLearningRateFinder(LearningRateFinder):
milestones = (1,)
def lr_find(self, trainer, pl_module) -> None:
pl_module._expected_max_steps = trainer.global_step + self._num_training_steps
super().lr_find(trainer, pl_module)
pl_module._expected_max_steps = None
assert not trainer.fit_loop.restarting
def on_train_epoch_start(self, trainer, pl_module):
if trainer.current_epoch in self.milestones or trainer.current_epoch == 0:
self.lr_find(trainer, pl_module)
model = MyBoringModel()
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=3,
callbacks=[
CustomLearningRateFinder(early_stop_threshold=None, update_attr=True, num_training_steps=num_lr_steps)
],
limit_train_batches=10,
limit_val_batches=0,
limit_test_batches=0,
num_sanity_val_steps=0,
enable_model_summary=False,
)
trainer.fit(model)
@mock.patch.dict(os.environ, os.environ.copy(), clear=True)
@RunIf(standalone=True)
def test_lr_finder_with_ddp(tmpdir):
seed_everything(7)
init_lr = 1e-4
dm = ClassifDataModule()
model = ClassificationModel(lr=init_lr)
trainer = Trainer(
default_root_dir=tmpdir,
max_epochs=1,
strategy="ddp",
devices=2,
accelerator="cpu",
)
tuner = Tuner(trainer)
tuner.lr_find(model, datamodule=dm, update_attr=True, num_training=20)
lr = trainer.lightning_module.lr
lr = trainer.strategy.broadcast(lr)
assert trainer.lightning_module.lr == lr
assert lr != init_lr
def test_lr_finder_callback_val_batches(tmpdir):
"""Test that `LearningRateFinder` does not limit the number of val batches during training."""
class CustomBoringModel(BoringModel):
def __init__(self, lr):
super().__init__()
self.lr = lr
def configure_optimizers(self):
return torch.optim.SGD(self.parameters(), lr=self.lr)
num_lr_tuner_training_steps = 5
model = CustomBoringModel(0.1)
trainer = Trainer(
default_root_dir=tmpdir,
num_sanity_val_steps=0,
max_epochs=1,
enable_model_summary=False,
callbacks=[LearningRateFinder(num_training_steps=num_lr_tuner_training_steps)],
)
trainer.fit(model)
assert trainer.num_val_batches[0] == len(trainer.val_dataloaders)
assert trainer.num_val_batches[0] != num_lr_tuner_training_steps
def test_lr_finder_training_step_none_output(tmpdir):
# add some nans into the skipped steps (first 10) but also into the steps used to compute the lr
none_steps = [5, 12, 17]
class CustomBoringModel(BoringModel):
def __init__(self):
super().__init__()
self.lr = 0.123
def training_step(self, batch: Any, batch_idx: int) -> STEP_OUTPUT:
if self.trainer.global_step in none_steps:
return None
return super().training_step(batch, batch_idx)
seed_everything(1)
model = CustomBoringModel()
trainer = Trainer(default_root_dir=tmpdir)
tuner = Tuner(trainer)
# restrict number of steps for faster test execution
# and disable early stopping to easily check expected number of lrs and losses
lr_finder = tuner.lr_find(model=model, update_attr=True, num_training=20, early_stop_threshold=None)
assert len(lr_finder.results["lr"]) == len(lr_finder.results["loss"]) == 20
assert torch.isnan(torch.tensor(lr_finder.results["loss"])[none_steps]).all()
suggested_lr = lr_finder.suggestion()
assert math.isfinite(suggested_lr)
assert math.isclose(model.lr, suggested_lr)