def find_learning_rate(self, train_data: TargetCollection,
results_dir: Path,
find_lr_kwargs: Optional[Dict[str, Any]] = None
) -> None:
'''
Given the training data it will plot learning rate against loss to allow
you to find the best learning rate.
This is just a wrapper around
allennlp.commands.find_learning_rate.find_learning_rate_model method.
:param train_data: Training data.
:param results_dir: Directory to store the results of the learning rate
findings.
:param find_lr_kwargs: Dictionary of keyword arguments to give to the
allennlp.commands.find_learning_rate.find_learning_rate_model
method.
'''
model_params = self._preprocess_and_load_param_file(self._param_fp)
with tempfile.TemporaryDirectory() as temp_dir:
train_fp = Path(temp_dir, 'train_data.json')
self._data_to_json(train_data, train_fp)
model_params['train_data_path'] = str(train_fp.resolve())
if find_lr_kwargs is None:
find_learning_rate_model(model_params, results_dir)
else:
find_learning_rate_model(model_params, results_dir, **find_lr_kwargs)
def test_find_learning_rate_multi_gpu(self):
params = self.params()
params["trainer"]["cuda_device"] = [0, 1]
find_learning_rate_model(params,
os.path.join(self.TEST_DIR, 'test_find_learning_rate_multi_gpu'),
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False)
def test_find_learning_rate_multi_gpu(self):
params = self.params()
params["trainer"]["cuda_device"] = [0, 1]
find_learning_rate_model(params,
os.path.join(
self.TEST_DIR,
'test_find_learning_rate_multi_gpu'),
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False)
def test_find_learning_rate(self):
find_learning_rate_model(
self.params(),
os.path.join(self.TEST_DIR, "test_find_learning_rate"),
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False,
)
# It's OK if serialization dir exists but is empty:
serialization_dir2 = os.path.join(self.TEST_DIR, "empty_directory")
assert not os.path.exists(serialization_dir2)
os.makedirs(serialization_dir2)
find_learning_rate_model(
self.params(),
serialization_dir2,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False,
)
# It's not OK if serialization dir exists and has junk in it non-empty:
serialization_dir3 = os.path.join(self.TEST_DIR, "non_empty_directory")
assert not os.path.exists(serialization_dir3)
os.makedirs(serialization_dir3)
with open(os.path.join(serialization_dir3, "README.md"), "w") as f:
f.write("TEST")
with pytest.raises(ConfigurationError):
find_learning_rate_model(
self.params(),
serialization_dir3,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False,
)
# ... unless you use the --force flag.
find_learning_rate_model(
self.params(),
serialization_dir3,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=True,
)
def test_find_learning_rate_multi_gpu(self):
params = self.params()
del params["trainer"]["cuda_device"]
params["distributed"] = Params({})
params["distributed"]["cuda_devices"] = [0, 1]
with pytest.raises(AssertionError) as execinfo:
find_learning_rate_model(
params,
os.path.join(self.TEST_DIR, "test_find_learning_rate_multi_gpu"),
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False,
)
assert "DistributedDataParallel" in str(execinfo.value)
def test_find_learning_rate(self):
find_learning_rate_model(self.params(),
os.path.join(self.TEST_DIR, 'test_find_learning_rate'),
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False)
# It's OK if serialization dir exists but is empty:
serialization_dir2 = os.path.join(self.TEST_DIR, 'empty_directory')
assert not os.path.exists(serialization_dir2)
os.makedirs(serialization_dir2)
find_learning_rate_model(self.params(), serialization_dir2,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False)
# It's not OK if serialization dir exists and has junk in it non-empty:
serialization_dir3 = os.path.join(self.TEST_DIR, 'non_empty_directory')
assert not os.path.exists(serialization_dir3)
os.makedirs(serialization_dir3)
with open(os.path.join(serialization_dir3, 'README.md'), 'w') as f:
f.write("TEST")
with pytest.raises(ConfigurationError):
find_learning_rate_model(self.params(), serialization_dir3,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=False)
# ... unless you use the --force flag.
find_learning_rate_model(self.params(), serialization_dir3,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None,
force=True)
def test_find_learning_rate(self):
find_learning_rate_model(self.params(),
serialization_dir=os.path.join(self.TEST_DIR, 'test_find_learning_rate'),
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None)
# It's OK if serialization dir exists but is empty:
serialization_dir2 = os.path.join(self.TEST_DIR, 'empty_directory')
assert not os.path.exists(serialization_dir2)
os.makedirs(serialization_dir2)
find_learning_rate_model(self.params(),
serialization_dir=serialization_dir2,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None)
# It's not OK if serialization dir exists and has junk in it non-empty:
serialization_dir3 = os.path.join(self.TEST_DIR, 'non_empty_directory')
assert not os.path.exists(serialization_dir3)
os.makedirs(serialization_dir3)
with open(os.path.join(serialization_dir3, 'README.md'), 'w') as f:
f.write("TEST")
with pytest.raises(ConfigurationError):
find_learning_rate_model(self.params(),
serialization_dir=serialization_dir3,
start_lr=1e-5,
end_lr=1,
num_batches=100,
linear_steps=True,
stopping_factor=None)
