def test_optimizer_return_options():
tutils.reset_seed()
trainer = Trainer()
model, hparams = tutils.get_default_model()
# single optimizer
opt_a = torch.optim.Adam(model.parameters(), lr=0.002)
opt_b = torch.optim.SGD(model.parameters(), lr=0.002)
optim, lr_sched = trainer.init_optimizers(opt_a)
assert len(optim) == 1 and len(lr_sched) == 0
# opt tuple
opts = (opt_a, opt_b)
optim, lr_sched = trainer.init_optimizers(opts)
assert len(optim) == 2 and optim[0] == opts[0] and optim[1] == opts[1]
assert len(lr_sched) == 0
# opt list
opts = [opt_a, opt_b]
optim, lr_sched = trainer.init_optimizers(opts)
assert len(optim) == 2 and optim[0] == opts[0] and optim[1] == opts[1]
assert len(lr_sched) == 0
# opt tuple of lists
scheduler = torch.optim.lr_scheduler.StepLR(opt_a, 10)
opts = ([opt_a], [scheduler])
optim, lr_sched = trainer.init_optimizers(opts)
assert len(optim) == 1 and len(lr_sched) == 1
assert optim[0] == opts[0][0] and \
lr_sched[0] == dict(scheduler=scheduler, interval='epoch',
frequency=1, reduce_on_plateau=False,
monitor='val_loss')
def test_adding_step_key(tmpdir):
logged_step = 0
def _validation_epoch_end(outputs):
nonlocal logged_step
logged_step += 1
return {"log": {"step": logged_step, "val_acc": logged_step / 10}}
def _training_epoch_end(outputs):
nonlocal logged_step
logged_step += 1
return {"log": {"step": logged_step, "train_acc": logged_step / 10}}
def _log_metrics_decorator(log_metrics_fn):
def decorated(metrics, step):
if "val_acc" in metrics:
assert step == logged_step
return log_metrics_fn(metrics, step)
return decorated
model, hparams = tutils.get_default_model()
model.validation_epoch_end = _validation_epoch_end
model.training_epoch_end = _training_epoch_end
trainer = Trainer(
max_epochs=4,
default_root_dir=tmpdir,
train_percent_check=0.001,
val_percent_check=0.01,
num_sanity_val_steps=0,
)
trainer.logger.log_metrics = _log_metrics_decorator(
trainer.logger.log_metrics)
trainer.fit(model)
def test_optimizer_return_options():
tutils.reset_seed()
trainer = Trainer()
model, hparams = tutils.get_default_model()
# single optimizer
opt_a = torch.optim.Adam(model.parameters(), lr=0.002)
opt_b = torch.optim.SGD(model.parameters(), lr=0.002)
scheduler_a = torch.optim.lr_scheduler.StepLR(opt_a, 10)
scheduler_b = torch.optim.lr_scheduler.StepLR(opt_b, 10)
# single optimizer
model.configure_optimizers = lambda: opt_a
optim, lr_sched, freq = trainer.init_optimizers(model)
assert len(optim) == 1 and len(lr_sched) == 0 and len(freq) == 0
# opt tuple
model.configure_optimizers = lambda: (opt_a, opt_b)
optim, lr_sched, freq = trainer.init_optimizers(model)
assert len(optim) == 2 and optim[0] == opt_a and optim[1] == opt_b
assert len(lr_sched) == 0 and len(freq) == 0
# opt list
model.configure_optimizers = lambda: [opt_a, opt_b]
optim, lr_sched, freq = trainer.init_optimizers(model)
assert len(optim) == 2 and optim[0] == opt_a and optim[1] == opt_b
assert len(lr_sched) == 0 and len(freq) == 0
# opt tuple of 2 lists
model.configure_optimizers = lambda: ([opt_a], [scheduler_a])
optim, lr_sched, freq = trainer.init_optimizers(model)
assert len(optim) == 1 and len(lr_sched) == 1 and len(freq) == 0
assert optim[0] == opt_a
assert lr_sched[0] == dict(scheduler=scheduler_a, interval='epoch',
frequency=1, reduce_on_plateau=False, monitor='val_loss')
# opt single dictionary
model.configure_optimizers = lambda: {"optimizer": opt_a, "lr_scheduler": scheduler_a}
optim, lr_sched, freq = trainer.init_optimizers(model)
assert len(optim) == 1 and len(lr_sched) == 1 and len(freq) == 0
assert optim[0] == opt_a
assert lr_sched[0] == dict(scheduler=scheduler_a, interval='epoch',
frequency=1, reduce_on_plateau=False, monitor='val_loss')
# opt multiple dictionaries with frequencies
model.configure_optimizers = lambda: (
{"optimizer": opt_a, "lr_scheduler": scheduler_a, "frequency": 1},
{"optimizer": opt_b, "lr_scheduler": scheduler_b, "frequency": 5},
)
optim, lr_sched, freq = trainer.init_optimizers(model)
assert len(optim) == 2 and len(lr_sched) == 2 and len(freq) == 2
assert optim[0] == opt_a
assert lr_sched[0] == dict(scheduler=scheduler_a, interval='epoch',
frequency=1, reduce_on_plateau=False, monitor='val_loss')
assert freq == [1, 5]
def test_none_optimizer_warning():
tutils.reset_seed()
trainer = Trainer()
model, hparams = tutils.get_default_model()
model.configure_optimizers = lambda: None
with pytest.warns(UserWarning, match='will run with no optimizer'):
_, __, ___ = trainer.init_optimizers(model)
def _init_steps_model():
"""private method for initializing a model with 5% train epochs"""
tutils.reset_seed()
model, _ = tutils.get_default_model()
# define train epoch to 5% of data
train_percent = 0.5
# get number of samples in 1 epoch
num_train_samples = math.floor(
len(model.train_dataloader()) * train_percent)
trainer_options = dict(train_percent_check=train_percent, )
return model, trainer_options, num_train_samples
def test_cpu_model(tmpdir):
"""Make sure model trains on CPU."""
tutils.reset_seed()
trainer_options = dict(default_root_dir=tmpdir,
progress_bar_refresh_rate=0,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.4)
model, hparams = tutils.get_default_model()
tutils.run_model_test(trainer_options, model, on_gpu=False)
def run_test_from_config(trainer_options):
"""Trains the default model with the given config."""
tutils.reset_seed()
tutils.set_random_master_port()
ckpt_path = trainer_options['default_root_dir']
trainer_options['checkpoint_callback'] = ModelCheckpoint(ckpt_path)
model, hparams = tutils.get_default_model()
tutils.run_model_test(trainer_options, model, version=0, with_hpc=False)
# Horovod should be initialized following training. If not, this will raise an exception.
assert hvd.size() == 2
def test_multi_gpu_none_backend(tmpdir):
"""Make sure when using multiple GPUs the user can't use `distributed_backend = None`."""
tutils.reset_seed()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_root_dir=tmpdir,
progress_bar_refresh_rate=0,
max_epochs=1,
train_percent_check=0.1,
val_percent_check=0.1,
gpus='-1')
with pytest.warns(UserWarning):
tutils.run_model_test(trainer_options, model)
def test_single_gpu_model(tmpdir):
"""Make sure single GPU works (DP mode)."""
tutils.reset_seed()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_root_dir=tmpdir,
progress_bar_refresh_rate=0,
max_epochs=1,
train_percent_check=0.1,
val_percent_check=0.1,
gpus=1)
tutils.run_model_test(trainer_options, model)
def test_cpu_model(tmpdir):
"""Make sure model trains on CPU."""
tutils.reset_seed()
trainer_options = dict(default_save_path=tmpdir,
show_progress_bar=False,
logger=tutils.get_default_testtube_logger(tmpdir),
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.4)
model, hparams = tutils.get_default_model()
tutils.run_model_test(trainer_options, model, on_gpu=False)
def test_cpu_model_with_amp(tmpdir):
"""Make sure model trains on CPU."""
tutils.reset_seed()
trainer_options = dict(default_root_dir=tmpdir,
progress_bar_refresh_rate=0,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.4,
precision=16)
model, hparams = tutils.get_default_model()
with pytest.raises((MisconfigurationException, ModuleNotFoundError)):
tutils.run_model_test(trainer_options, model, on_gpu=False)
def test_lbfgs_cpu_model(tmpdir):
"""Test each of the trainer options."""
tutils.reset_seed()
trainer_options = dict(
default_root_dir=tmpdir,
max_epochs=2,
progress_bar_refresh_rate=0,
weights_summary='top',
train_percent_check=1.0,
val_percent_check=0.2,
)
model, hparams = tutils.get_default_model(lbfgs=True)
tutils.run_model_test_no_loggers(trainer_options, model, min_acc=0.5)
def test_multi_gpu_model_ddp(tmpdir):
"""Make sure DDP works."""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_save_path=tmpdir,
show_progress_bar=False,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.2,
gpus=[0, 1],
distributed_backend='ddp')
tutils.run_model_test(trainer_options, model)
def test_all_features_cpu_model(tmpdir):
"""Test each of the trainer options."""
tutils.reset_seed()
trainer_options = dict(default_root_dir=tmpdir,
gradient_clip_val=1.0,
overfit_pct=0.20,
track_grad_norm=2,
progress_bar_refresh_rate=0,
accumulate_grad_batches=2,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.4)
model, hparams = tutils.get_default_model()
tutils.run_model_test(trainer_options, model, on_gpu=False)
def test_multi_cpu_model_ddp(tmpdir):
"""Make sure DDP works."""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_root_dir=tmpdir,
progress_bar_refresh_rate=0,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.2,
gpus=None,
num_processes=2,
distributed_backend='ddp_cpu')
tutils.run_model_test(trainer_options, model, on_gpu=False)
def test_amp_gpu_dp(tmpdir):
"""Make sure DP + AMP work."""
tutils.reset_seed()
model, hparams = tutils.get_default_model()
trainer_options = dict(
default_save_path=tmpdir,
max_epochs=1,
gpus='0, 1', # test init with gpu string
distributed_backend='dp',
precision=16)
trainer = Trainer(**trainer_options)
result = trainer.fit(model)
assert result == 1
def test_multi_gpu_model_ddp2(tmpdir):
"""Make sure DDP2 works."""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_save_path=tmpdir,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.2,
gpus=2,
weights_summary=None,
distributed_backend='ddp2')
tutils.run_model_test(trainer_options, model)
def test_multi_gpu_model_dp(tmpdir):
"""Make sure DP works."""
tutils.reset_seed()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_save_path=tmpdir,
show_progress_bar=False,
distributed_backend='dp',
max_epochs=1,
train_percent_check=0.1,
val_percent_check=0.1,
gpus='-1')
tutils.run_model_test(trainer_options, model)
# test memory helper functions
memory.get_memory_profile('min_max')
def test_multi_gpu_none_backend(tmpdir):
"""Make sure when using multiple GPUs the user can't use `distributed_backend = None`."""
tutils.reset_seed()
if not tutils.can_run_gpu_test():
return
model, hparams = tutils.get_default_model()
trainer_options = dict(default_save_path=tmpdir,
show_progress_bar=False,
max_epochs=1,
train_percent_check=0.1,
val_percent_check=0.1,
gpus='-1')
with pytest.warns(UserWarning):
tutils.run_model_test(trainer_options, model)
def test_single_gpu_model(tmpdir):
"""Make sure single GPU works (DP mode)."""
tutils.reset_seed()
if not torch.cuda.is_available():
warnings.warn('test_single_gpu_model cannot run.'
' Rerun on a GPU node to run this test')
return
model, hparams = tutils.get_default_model()
trainer_options = dict(default_save_path=tmpdir,
show_progress_bar=False,
max_epochs=1,
train_percent_check=0.1,
val_percent_check=0.1,
gpus=1)
tutils.run_model_test(trainer_options, model)
def test_amp_single_gpu(tmpdir, backend):
"""Make sure DP/DDP + AMP work."""
tutils.reset_seed()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_root_dir=tmpdir,
max_epochs=1,
gpus=1,
distributed_backend=backend,
precision=16)
# tutils.run_model_test(trainer_options, model)
trainer = Trainer(**trainer_options)
result = trainer.fit(model)
assert result == 1
def test_amp_multi_gpu(tmpdir, backend):
"""Make sure DP/DDP + AMP work."""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer_options = dict(
default_root_dir=tmpdir,
max_epochs=1,
# gpus=2,
gpus='0, 1', # test init with gpu string
distributed_backend=backend,
precision=16)
# tutils.run_model_test(trainer_options, model)
trainer = Trainer(**trainer_options)
result = trainer.fit(model)
assert result
def test_ddp_all_dataloaders_passed_to_fit(tmpdir):
"""Make sure DDP works with dataloaders passed to fit()"""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer = Trainer(default_root_dir=tmpdir,
progress_bar_refresh_rate=0,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.2,
gpus=[0, 1],
distributed_backend='ddp')
result = trainer.fit(model,
train_dataloader=model.train_dataloader(),
val_dataloaders=model.val_dataloader())
assert result == 1, "DDP doesn't work with dataloaders passed to fit()."
def test_default_logger_callbacks_cpu_model(tmpdir):
"""Test each of the trainer options."""
tutils.reset_seed()
trainer_options = dict(
default_root_dir=tmpdir,
max_epochs=1,
gradient_clip_val=1.0,
overfit_pct=0.20,
progress_bar_refresh_rate=0,
train_percent_check=0.01,
val_percent_check=0.01,
)
model, hparams = tutils.get_default_model()
tutils.run_model_test_no_loggers(trainer_options, model)
# test freeze on cpu
model.freeze()
model.unfreeze()
def test_ddp_all_dataloaders_passed_to_fit(tmpdir):
"""Make sure DDP works with dataloaders passed to fit()"""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer_options = dict(default_save_path=tmpdir,
show_progress_bar=False,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.2,
gpus=[0, 1],
distributed_backend='ddp')
fit_options = dict(train_dataloader=model.train_dataloader(),
val_dataloaders=model.val_dataloader())
trainer = Trainer(**trainer_options)
result = trainer.fit(model, **fit_options)
assert result == 1, "DDP doesn't work with dataloaders passed to fit()."
def test_early_stopping_cpu_model(tmpdir):
"""Test each of the trainer options."""
tutils.reset_seed()
stopping = EarlyStopping(monitor='val_loss', min_delta=0.1)
trainer_options = dict(
default_root_dir=tmpdir,
early_stop_callback=stopping,
gradient_clip_val=1.0,
overfit_pct=0.20,
track_grad_norm=2,
train_percent_check=0.1,
val_percent_check=0.1,
)
model, hparams = tutils.get_default_model()
tutils.run_model_test(trainer_options, model, on_gpu=False)
# test freeze on cpu
model.freeze()
model.unfreeze()
def test_loggers_fit_test(tmpdir, monkeypatch, logger_class):
"""Verify that basic functionality of all loggers."""
tutils.reset_seed()
# prevent comet logger from trying to print at exit, since
# pytest's stdout/stderr redirection breaks it
import atexit
monkeypatch.setattr(atexit, 'register', lambda _: None)
model, _ = tutils.get_default_model()
class StoreHistoryLogger(logger_class):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.history = []
def log_metrics(self, metrics, step):
super().log_metrics(metrics, step)
self.history.append((step, metrics))
logger_args = _get_logger_args(logger_class, tmpdir)
logger = StoreHistoryLogger(**logger_args)
trainer = Trainer(
max_epochs=1,
logger=logger,
train_percent_check=0.2,
val_percent_check=0.5,
fast_dev_run=True,
)
trainer.fit(model)
trainer.test()
log_metric_names = [(s, sorted(m.keys())) for s, m in logger.history]
assert log_metric_names == [(0, ['val_acc', 'val_loss']),
(0, ['train_some_val']),
(1, ['test_acc', 'test_loss'])]
def test_multi_gpu_model(tmpdir, backend):
"""Make sure DDP works."""
tutils.reset_seed()
tutils.set_random_master_port()
model, hparams = tutils.get_default_model()
trainer_options = dict(
default_root_dir=tmpdir,
max_epochs=1,
train_percent_check=0.4,
val_percent_check=0.2,
gpus=[0, 1],
distributed_backend=backend,
)
# tutils.run_model_test(trainer_options, model)
trainer = Trainer(**trainer_options)
result = trainer.fit(model)
assert result
# test memory helper functions
memory.get_memory_profile('min_max')