def test_default():
runner = Runner(model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=None)
runner.fit(TestLoader, epochs=2)
def test_default():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=None,
)
runner.fit(TEST_LOADER, epochs=2)
def test_accumulate_steps():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
accumulate_steps=10,
)
runner.fit(TEST_LOADER, epochs=2)
def test_segm_callback(callback):
runner = Runner(
model=TEST_SEGM_MODEL,
optimizer=TEST_SEGM_OPTIMZER,
criterion=TEST_CRITERION,
callbacks=callback,
)
runner.fit(TEST_SEGM_LOADER, epochs=2)
def test_callback(callback):
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=[callback, pt_clb.BatchMetrics(TEST_METRIC)],
)
runner.fit(TEST_LOADER, epochs=2)
def test_ModelEma_callback():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=pt_clb.ModelEma(TEST_MODEL),
)
runner.fit(TEST_LOADER, epochs=2)
def test_fp16_training():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
use_fp16=True,
)
runner.fit(TEST_LOADER, epochs=2)
def test_grad_clip_loader():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
gradient_clip_val=1.0,
)
runner.fit(TEST_LOADER, epochs=2)
def test_accumulate_steps_with_clip_grad():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=[pt_clb.GradientClipping(1)],
accumulate_steps=4,
)
runner.fit(TEST_LOADER, epochs=2)
def test_Timer_callback():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=pt_clb.Timer(),
)
runner.fit(TestLoader, epochs=2)
def test_Mixup():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=pt_clb.Mixup(0.2, NUM_CLASSES),
)
runner.fit(TestLoader, epochs=2)
def test_val_loader():
runner = Runner(model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION)
runner.fit(TEST_LOADER,
epochs=2,
steps_per_epoch=100,
val_loader=TEST_LOADER,
val_steps=200)
def test_FileLogger_callback():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=pt_clb.FileLogger(TMP_PATH),
)
runner.fit(TestLoader, epochs=2)
def test_TensorBoard():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=pt_clb.TensorBoard(log_dir=TMP_PATH),
)
runner.fit(TestLoader, epochs=2)
def test_ReduceLROnPlateau_callback():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=pt_clb.ReduceLROnPlateau(),
)
runner.fit(TestLoader, epochs=2)
def test_CheckpointSaver_callback():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
callbacks=pt_clb.CheckpointSaver(TMP_PATH, save_name="model.chpn"),
)
runner.fit(TestLoader, epochs=2)
def test_callback(callback):
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
metrics=TEST_METRIC,
callbacks=callback,
)
runner.fit(TEST_LOADER, epochs=2)
def test_tensorboar_CM():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=[
pt_clb.TensorBoardCM(),
pt_clb.TensorBoard(log_dir=TMP_PATH)
],
)
runner.fit(TEST_LOADER, epochs=2)
def test_val_loader():
runner = Runner(
model=TestModel,
optimizer=TestOptimizer,
criterion=TestCriterion,
metrics=TestMetric,
)
runner.fit(TestLoader,
epochs=2,
steps_per_epoch=100,
val_loader=TestLoader,
val_steps=200)
def test_loader_metric():
"""Check that LoaderMetric doesn't store grads and results are on cpu to avoid memory leak"""
clb = pt_clb.LoaderMetrics(TEST_METRIC)
runner = Runner(model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=clb)
runner.fit(TEST_LOADER, epochs=2)
assert clb.target[0].grad_fn is None
assert clb.output[0].grad_fn is None
assert clb.target[0].device == torch.device("cpu")
assert clb.output[0].device == torch.device("cpu")
def test_invalid_phases_scheduler_mode():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=pt_clb.PhasesScheduler([
{
"ep": [0, 1],
"lr": [0, 1],
"mode": "new_mode"
},
]),
)
with pytest.raises(ValueError):
runner.fit(TEST_LOADER, epochs=2)
def test_state_is_frozen():
runner = Runner(model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=None)
with pytest.raises(TypeError):
setattr(runner.state, "something", "value")
def test_state_batch_size():
runner = Runner(
model=TEST_MODEL,
optimizer=TEST_OPTIMIZER,
criterion=TEST_CRITERION,
callbacks=None,
)
runner.fit(TEST_LOADER, epochs=1)
# check that batch_size is copied correctly
assert runner.state.batch_size == BS
# check that if batch_size is not given, it would be 1
loader = deepcopy(TEST_LOADER)
delattr(loader, "batch_size")
runner.fit(loader, epochs=1)
assert runner.state.batch_size == 1
