def test_structured_data_input_less_col_name_error():
with pytest.raises(ValueError) as info:
analyser = input_analysers.StructuredDataAnalyser(
column_names=list(range(8)))
dataset = tf.data.Dataset.from_tensor_slices(np.random.rand(
20, 10)).batch(32)
for x in dataset:
analyser.update(x)
analyser.finalize()
assert "Expect column_names to have length" in str(info.value)
def test_structured_data_input_with_illegal_dim():
analyser = input_analysers.StructuredDataAnalyser(
column_names=test_utils.COLUMN_NAMES,
column_types=None,
)
dataset = tf.data.Dataset.from_tensor_slices(np.random.rand(100, 32, 32)).batch(
32
)
with pytest.raises(ValueError) as info:
for data in dataset:
analyser.update(data)
analyser.finalize()
assert "Expect the data to StructuredDataInput to have shape" in str(info.value)
def test_structured_data_infer_col_types():
analyser = input_analysers.StructuredDataAnalyser(
column_names=utils.COLUMN_NAMES,
column_types=None,
)
x = pd.read_csv(utils.TRAIN_CSV_PATH)
x.pop("survived")
dataset = tf.data.Dataset.from_tensor_slices(x.values.astype(
np.unicode)).batch(32)
for data in dataset:
analyser.update(data)
analyser.finalize()
assert analyser.column_types == utils.COLUMN_TYPES
def test_dont_infer_specified_column_types():
column_types = copy.copy(utils.COLUMN_TYPES)
column_types.pop("sex")
column_types["age"] = "categorical"
analyser = input_analysers.StructuredDataAnalyser(
column_names=utils.COLUMN_NAMES,
column_types=column_types,
)
x = pd.read_csv(utils.TRAIN_CSV_PATH)
x.pop("survived")
dataset = tf.data.Dataset.from_tensor_slices(x.values.astype(
np.unicode)).batch(32)
for data in dataset:
analyser.update(data)
analyser.finalize()
assert analyser.column_types["age"] == "categorical"