def test_create_float_list_tensor(self):
tensorizer = FloatListTensorizer(
column="dense", dim=2, error_check=True, normalize=False
)
tests = [
("[0.1,0.2]", [0.1, 0.2]), # comma
("[0.1, 0.2]", [0.1, 0.2]), # comma with single space
("[0.1, 0.2]", [0.1, 0.2]), # comma with multiple spaces
("[0.1 0.2]", [0.1, 0.2]), # space
("[0.1 0.2]", [0.1, 0.2]), # multiple spaces
("[ 0.1 0.2]", [0.1, 0.2]), # space after [
("[0.1 0.2 ]", [0.1, 0.2]), # space before ]
("[0. 1.]", [0.0, 1.0]), # 0., 1.
]
for raw, expected in tests:
row = {"dense": load_float_list(raw)}
numberized = tensorizer.numberize(row)
self.assertEqual(expected, numberized)
precision.FP16_ENABLED = True
batch = []
for raw, _ in tests:
row = {"dense": load_float_list(raw)}
batch.append(tensorizer.numberize(row))
tensor = tensorizer.tensorize(batch)
self.assertEqual(list(tensor.size()), [8, 2])
self.assertEqual(tensor.dtype, torch.float16)
precision.FP16_ENABLED = False
def test_float_list_tensor_prepare_input(self):
tensorizer = FloatListTensorizer(
column="dense", dim=2, error_check=True, normalize=False
)
tests = [("[0.1,0.2]", [0.1, 0.2])]
for raw, expected in tests:
row = {"dense": load_float_list(raw)}
numberized = tensorizer.prepare_input(row)
self.assertEqual(expected, numberized)
def test_create_float_list_tensor(self):
tensorizer = FloatListTensorizer(column="dense", dim=2, error_check=True)
rows = [
{"dense": "[0.1,0.2]"}, # comma
{"dense": "[0.1, 0.2]"}, # comma with single space
{"dense": "[0.1, 0.2]"}, # comma with multiple spaces
{"dense": "[0.1 0.2]"}, # space
{"dense": "[0.1 0.2]"}, # multiple spaces
{"dense": "[ 0.1 0.2]"}, # space after [
{"dense": "[0.1 0.2 ]"}, # space before ]
]
tensors = (tensorizer.numberize(row) for row in rows)
for tensor in tensors:
self.assertEqual([0.1, 0.2], tensor)
def _get_config_with_export_list(
self,
task_class: Type[NewTask],
model_class: Type[Model],
test_file_metadata: TestFileMetadata,
) -> PyTextConfig:
return PyTextConfig(
task=task_class.Config(
data=Data.Config(
source=TSVDataSource.Config(
train_filename=test_file_metadata.filename,
eval_filename=test_file_metadata.filename,
test_filename=test_file_metadata.filename,
field_names=test_file_metadata.field_names,
),
batcher=PoolingBatcher.Config(train_batch_size=1,
test_batch_size=1),
),
trainer=TaskTrainer.Config(epochs=1),
model=model_class.Config(
inputs=type(model_class.Config.inputs)(
dense=FloatListTensorizer.Config(
column=test_file_metadata.dense_col_name,
error_check=True,
dim=test_file_metadata.dense_feat_dim,
))),
),
use_tensorboard=False,
use_cuda_if_available=False,
export=ExportConfig(
export_torchscript_path="/tmp/model_torchscript.pt"),
version=LATEST_VERSION,
)
def _get_pytext_config(
self,
test_file_name: TestFileName,
task_class: Type[NewTask],
model_class: Type[Model],
) -> PyTextConfig:
test_file_metadata = get_test_file_metadata(test_file_name)
return PyTextConfig(
task=task_class.Config(
data=Data.Config(
source=TSVDataSource.Config(
train_filename=test_file_metadata.filename,
eval_filename=test_file_metadata.filename,
test_filename=test_file_metadata.filename,
field_names=test_file_metadata.field_names,
),
batcher=Batcher.Config(
), # Use Batcher to avoid shuffling.
),
trainer=TaskTrainer.Config(epochs=1),
model=model_class.Config(
inputs=type(model_class.Config.inputs)(
dense=FloatListTensorizer.Config(
column=test_file_metadata.dense_col_name,
dim=test_file_metadata.dense_feat_dim,
))),
),
use_tensorboard=False,
use_cuda_if_available=False,
version=LATEST_VERSION,
)
def test_create_float_list_tensor(self):
tensorizer = FloatListTensorizer(column="dense", dim=2, error_check=True)
tests = [
("[0.1,0.2]", [0.1, 0.2]), # comma
("[0.1, 0.2]", [0.1, 0.2]), # comma with single space
("[0.1, 0.2]", [0.1, 0.2]), # comma with multiple spaces
("[0.1 0.2]", [0.1, 0.2]), # space
("[0.1 0.2]", [0.1, 0.2]), # multiple spaces
("[ 0.1 0.2]", [0.1, 0.2]), # space after [
("[0.1 0.2 ]", [0.1, 0.2]), # space before ]
("[0. 1.]", [0.0, 1.0]), # 0., 1.
]
for raw, expected in tests:
row = {"dense": load_float_list(raw)}
numberized = tensorizer.numberize(row)
self.assertEqual(expected, numberized)
def test_batch_predict_caffe2_model(self):
with tempfile.NamedTemporaryFile() as snapshot_file, tempfile.NamedTemporaryFile() as caffe2_model_file:
train_data = tests_module.test_file("train_data_tiny.tsv")
eval_data = tests_module.test_file("test_data_tiny.tsv")
config = PyTextConfig(
task=DocumentClassificationTask.Config(
model=DocModel.Config(
inputs=DocModel.Config.ModelInput(
tokens=TokenTensorizer.Config(),
dense=FloatListTensorizer.Config(
column="dense", dim=1, error_check=True
),
labels=LabelTensorizer.Config(),
)
),
data=Data.Config(
source=TSVDataSource.Config(
train_filename=train_data,
eval_filename=eval_data,
test_filename=eval_data,
field_names=["label", "slots", "text", "dense"],
)
),
),
version=21,
save_snapshot_path=snapshot_file.name,
export_caffe2_path=caffe2_model_file.name,
)
task = create_task(config.task)
task.export(task.model, caffe2_model_file.name)
model = task.model
save(config, model, meta=None, tensorizers=task.data.tensorizers)
pt_results = task.predict(task.data.data_source.test)
def assert_caffe2_results_correct(caffe2_results):
for pt_res, res in zip(pt_results, caffe2_results):
np.testing.assert_array_almost_equal(
pt_res["score"].tolist()[0],
[score[0] for score in res.values()],
)
results = batch_predict_caffe2_model(
snapshot_file.name, caffe2_model_file.name
)
self.assertEqual(4, len(results))
assert_caffe2_results_correct(results)
results = batch_predict_caffe2_model(
snapshot_file.name, caffe2_model_file.name, cache_size=2
)
self.assertEqual(4, len(results))
assert_caffe2_results_correct(results)
results = batch_predict_caffe2_model(
snapshot_file.name, caffe2_model_file.name, cache_size=-1
)
self.assertEqual(4, len(results))
assert_caffe2_results_correct(results)
def test_create_normalized_float_list_tensor(self):
def round_list(l):
return [float("%.4f" % n) for n in l]
data = TSVDataSource(
SafeFileWrapper(tests_module.test_file("train_dense_features_tiny.tsv")),
eval_file=None,
field_names=["label", "slots", "text", "dense_feat"],
schema={"text": str, "label": str, "dense_feat": List[float]},
)
tensorizer = FloatListTensorizer(
column="dense_feat", dim=10, error_check=True, normalize=True
)
self._initialize_tensorizer(tensorizer, data)
self.assertEqual(10, tensorizer.normalizer.num_rows)
self.assertEqual(
round_list(
[
7.56409,
8.2388,
0.5531,
0.2403,
1.03130,
6.2888,
3.1595,
0.1538,
0.2403,
5.3463,
]
),
round_list(tensorizer.normalizer.feature_sums),
)
self.assertEqual(
round_list(
[
5.80172,
7.57586,
0.30591,
0.05774,
0.52762,
5.22811,
2.51727,
0.02365,
0.05774,
4.48798,
]
),
round_list(tensorizer.normalizer.feature_squared_sums),
)
self.assertEqual(
round_list(
[
0.75640,
0.82388,
0.05531,
0.02403,
0.10313,
0.62888,
0.31595,
0.01538,
0.02403,
0.53463,
]
),
round_list(tensorizer.normalizer.feature_avgs),
)
self.assertEqual(
round_list(
[
0.08953,
0.28072,
0.16593,
0.07209,
0.20524,
0.35682,
0.38974,
0.04614,
0.07209,
0.40369,
]
),
round_list(tensorizer.normalizer.feature_stddevs),
)
row = [0.64840776, 0.7575, 0.5531, 0.2403, 0, 0.9481, 0, 0.1538, 0.2403, 0.3564]
output = tensorizer.numberize({"dense_feat": row})
self.assertEqual(
round_list(
[
-1.20619,
-0.23646,
2.99999,
3.0,
-0.50246,
0.89462,
-0.81066,
2.99999,
3.0,
-0.44149,
]
),
round_list(output),
)
