def output_spec(self) -> lit_types.Spec:
"""Return a spec describing model outputs."""
output_spec_dict = dict(self._output_types)
if self.attribution_explainer:
output_spec_dict[
"feature_attribution"] = lit_types.FeatureSalience(signed=True)
return output_spec_dict
def test_set_up_and_open_lit_with_xai(self, set_up_sequential,
set_up_pandas_dataframe_and_columns,
widget_render_mock):
pd_dataset, lit_columns = set_up_pandas_dataframe_and_columns
feature_types, label_types, saved_model_path = set_up_sequential
lit_dataset, lit_model = set_up_and_open_lit(pd_dataset, lit_columns,
saved_model_path,
feature_types,
label_types)
expected_examples = [
{
"feature_1": 1.0,
"feature_2": 3.0,
"label": 1.0
},
]
test_inputs = [
{
"feature_1": 1.0,
"feature_2": 2.0
},
]
outputs = lit_model.predict_minibatch(test_inputs)
assert lit_dataset.spec() == dict(lit_columns)
assert expected_examples == lit_dataset._examples
assert lit_model.input_spec() == dict(feature_types)
assert lit_model.output_spec() == dict({
**label_types,
"feature_attribution":
lit_types.FeatureSalience(signed=True),
})
assert len(outputs) == 1
for item in outputs:
assert item.keys() == {"label", "feature_attribution"}
assert len(item.values()) == 2
widget_render_mock.assert_called_once()
def test_create_lit_model_from_endpoint_name_with_xai_returns_model(
self, feature_types, label_types, model_id):
lit_model = create_lit_model_from_endpoint(_TEST_ENDPOINT_NAME,
feature_types, label_types,
model_id)
test_inputs = [
{
"feature_1": 1.0,
"feature_2": 2.0
},
]
outputs = lit_model.predict_minibatch(test_inputs)
assert lit_model.input_spec() == dict(feature_types)
assert lit_model.output_spec() == dict({
**label_types,
"feature_attribution":
lit_types.FeatureSalience(signed=True),
})
assert len(outputs) == 1
for item in outputs:
assert item.keys() == {"label", "feature_attribution"}
assert len(item.values()) == 2
def test_create_lit_model_from_tensorflow_with_xai_returns_model(
self, set_up_sequential):
feature_types, label_types, saved_model_path = set_up_sequential
lit_model = create_lit_model(saved_model_path, feature_types,
label_types)
test_inputs = [
{
"feature_1": 1.0,
"feature_2": 2.0
},
]
outputs = lit_model.predict_minibatch(test_inputs)
assert lit_model.input_spec() == dict(feature_types)
assert lit_model.output_spec() == dict({
**label_types,
"feature_attribution":
lit_types.FeatureSalience(signed=True),
})
assert len(outputs) == 1
for item in outputs:
assert item.keys() == {"label", "feature_attribution"}
assert len(item.values()) == 2
