def test_pipeline_postprocess_multiple_one_hot_to_labels():
pipeline = pipeline_module.Pipeline(
inputs=[[]],
outputs=[
[preprocessors.OneHotEncoder(["a", "b", "c"])],
[preprocessors.OneHotEncoder(["a", "b", "c"])],
],
)
result = pipeline.postprocess([np.eye(3), np.eye(3)])
assert np.array_equal(result[0], [["a"], ["b"], ["c"]])
assert np.array_equal(result[1], [["a"], ["b"], ["c"]])
def get_hyper_preprocessors(self):
hyper_preprocessors = []
if self._add_one_dimension:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.AddOneDimension()))
if self._dtype in [tf.uint8, tf.uint16, tf.uint32, tf.uint64]:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.CastToInt32()))
if not self._encoded and self._dtype != tf.string:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.CastToString()))
if self.multi_label:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.MultiLabelEncoder()))
if not self._encoded:
if self.num_classes == 2 and not self.multi_label:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.LabelEncoder(self._labels)))
else:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.OneHotEncoder(self._labels)))
return hyper_preprocessors
def get_hyper_preprocessors(self):
hyper_preprocessors = []
if self._add_one_dimension:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.AddOneDimension()))
if self.dtype in [tf.uint8, tf.uint16, tf.uint32, tf.uint64]:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.CastToInt32()))
if not self._encoded and self.dtype != tf.string:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.CastToString()))
if self._encoded_for_sigmoid:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.SigmoidPostprocessor()))
elif self._encoded_for_softmax:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.SoftmaxPostprocessor()))
elif self.num_classes == 2:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.LabelEncoder(self._labels)))
else:
hyper_preprocessors.append(
hpps_module.DefaultHyperPreprocessor(
preprocessors.OneHotEncoder(self._labels)))
return hyper_preprocessors
def test_pipeline_postprocess_one_hot_to_labels():
pipeline = pipeline_module.Pipeline(
inputs=[[]], outputs=[[preprocessors.OneHotEncoder(["a", "b", "c"])]])
assert np.array_equal(pipeline.postprocess(np.eye(3)),
[["a"], ["b"], ["c"]])