def test_csv_decoder_with_schema(self):
input_lines = ['1,1,2.0,hello', '5,5,12.34,world']
column_names = [
'int_feature_parsed_as_float', 'int_feature', 'float_feature',
'str_feature'
]
schema = text_format.Parse(
"""
feature { name: "int_feature_parsed_as_float" type: FLOAT }
feature { name: "int_feature" type: INT }
feature { name: "float_feature" type: FLOAT }
feature { name: "str_feature" type: BYTES }
""", schema_pb2.Schema())
expected_result = [
pa.Table.from_arrays([
pa.array([[1], [5]], pa.list_(pa.float32())),
pa.array([[1], [5]], pa.list_(pa.int64())),
pa.array([[2.0], [12.34]], pa.list_(pa.float32())),
pa.array([[b'hello'], [b'world']], pa.list_(pa.binary())),
], [
'int_feature_parsed_as_float', 'int_feature', 'float_feature',
'str_feature'
])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines)
| csv_decoder.DecodeCSV(column_names=column_names,
schema=schema,
infer_type_from_schema=True))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder_with_float_and_string_in_same_column(self):
input_lines = ['2.3,abc', 'abc,2.3']
column_names = ['str_feature1', 'str_feature2']
expected_result = [
pa.Table.from_arrays([
pa.array([[b'2.3'], [b'abc']], pa.list_(pa.binary())),
pa.array([[b'abc'], [b'2.3']], pa.list_(pa.binary())),
], ['str_feature1', 'str_feature2'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines)
| csv_decoder.DecodeCSV(column_names=column_names))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_topk_uniques_combiner_zero_row(self):
batches = [
pa.Table.from_arrays([pa.array([], type=pa.list_(pa.binary()))],
['f1'])
]
expected_result = {}
generator = (top_k_uniques_combiner_stats_generator.
TopKUniquesCombinerStatsGenerator(
num_top_values=4, num_rank_histogram_buckets=3))
self.assertCombinerOutputEqual(batches, generator, expected_result)
def test_get_binary_array_total_byte_size(self):
binary_array = pa.array([b"abc", None, b"def", b"", b"ghi"])
self.assertEqual(9, arrow_util.GetBinaryArrayTotalByteSize(binary_array))
sliced_1_2 = binary_array.slice(1, 2)
self.assertEqual(3, arrow_util.GetBinaryArrayTotalByteSize(sliced_1_2))
sliced_2 = binary_array.slice(2)
self.assertEqual(6, arrow_util.GetBinaryArrayTotalByteSize(sliced_2))
unicode_array = pa.array([u"abc"])
self.assertEqual(3, arrow_util.GetBinaryArrayTotalByteSize(unicode_array))
empty_array = pa.array([], type=pa.binary())
self.assertEqual(0, arrow_util.GetBinaryArrayTotalByteSize(empty_array))
def test_csv_decoder_with_unicode(self):
input_lines = [u'1,שקרכלשהו,22.34,text field']
column_names = [
'int_feature', 'unicode_feature', 'float_feature', 'str_feature'
]
expected_result = [
pa.Table.from_arrays([
pa.array([[1]], pa.list_(pa.int64())),
pa.array([[22.34]], pa.list_(pa.float32())),
pa.array([[u'שקרכלשהו'.encode('utf-8')]], pa.list_(
pa.binary())),
pa.array([[b'text field']], pa.list_(pa.binary())),
], [
'int_feature', 'float_feature', 'unicode_feature',
'str_feature'
])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines)
| csv_decoder.DecodeCSV(column_names=column_names))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder_large_int_categorical_neg(self):
input_lines = ['34', str(-(sys.maxsize + 2))]
column_names = ['feature']
expected_result = [
pa.Table.from_arrays([
pa.array([[b'34'], [str(-(sys.maxsize + 2)).encode('utf-8')]],
pa.list_(pa.binary())),
], ['feature'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines)
| csv_decoder.DecodeCSV(column_names=column_names))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder_with_tab_delimiter(self):
input_lines = ['1\t"this is a \ttext"', '5\t']
column_names = ['int_feature', 'str_feature']
expected_result = [
pa.Table.from_arrays([
pa.array([[1], [5]], pa.list_(pa.int64())),
pa.array([[b'this is a \ttext'], None], pa.list_(pa.binary())),
], ['int_feature', 'str_feature'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines) | csv_decoder.DecodeCSV(
column_names=column_names, delimiter='\t'))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder_with_space_delimiter(self):
input_lines = ['1 "ab,cd,ef"', '5 "wx,xy,yz"']
column_names = ['int_feature', 'str_feature']
expected_result = [
pa.Table.from_arrays([
pa.array([[1], [5]], pa.list_(pa.int64())),
pa.array([[b'ab,cd,ef'], [b'wx,xy,yz']], pa.list_(
pa.binary())),
], ['int_feature', 'str_feature'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines) | csv_decoder.DecodeCSV(
column_names=column_names, delimiter=' '))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder_empty_row(self):
input_lines = [',,', '1,2.0,hello']
column_names = ['int_feature', 'float_feature', 'str_feature']
expected_result = [
pa.Table.from_arrays([
pa.array([None, [1.0]], pa.list_(pa.float32())),
pa.array([None, [2.0]], pa.list_(pa.float32())),
pa.array([None, [b'hello']], pa.list_(pa.binary())),
], ['int_feature', 'float_feature', 'str_feature'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines)
| csv_decoder.DecodeCSV(column_names=column_names))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder(self):
input_lines = ['1,2.0,hello', '5,12.34,world']
column_names = ['int_feature', 'float_feature', 'str_feature']
expected_result = [
pa.Table.from_arrays([
pa.array([[1], [5]], pa.list_(pa.int64())),
pa.array([[2.0], [12.34]], pa.list_(pa.float32())),
pa.array([[b'hello'], [b'world']], pa.list_(pa.binary())),
], ['int_feature', 'float_feature', 'str_feature'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines, reshuffle=False)
| csv_decoder.DecodeCSV(column_names=column_names))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def test_csv_decoder_missing_values(self):
input_lines = ['1,,hello', ',12.34,']
column_names = ['int_feature', 'float_feature', 'str_feature']
expected_result = [
pa.Table.from_arrays([
pa.array([[1], None], pa.list_(pa.int64())),
pa.array([None, [12.34]], pa.list_(pa.float32())),
pa.array([[b'hello'], None], pa.list_(pa.binary())),
], ['int_feature', 'float_feature', 'str_feature'])
]
with beam.Pipeline() as p:
result = (p | beam.Create(input_lines)
| csv_decoder.DecodeCSV(column_names=column_names))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_result))
def _process_column_infos(self, column_infos: List[csv_decoder.ColumnInfo]):
column_handlers = []
column_arrow_types = []
for c in column_infos:
if c.type == statistics_pb2.FeatureNameStatistics.INT:
column_handlers.append(lambda v: (int(v),))
column_arrow_types.append(pa.list_(pa.int64()))
elif c.type == statistics_pb2.FeatureNameStatistics.FLOAT:
column_handlers.append(lambda v: (float(v),))
column_arrow_types.append(pa.list_(pa.float32()))
elif c.type == statistics_pb2.FeatureNameStatistics.STRING:
column_handlers.append(lambda v: (v,))
column_arrow_types.append(pa.list_(pa.binary()))
else:
column_handlers.append(lambda _: None)
column_arrow_types.append(pa.null())
self._column_handlers = column_handlers
self._column_arrow_types = column_arrow_types
self._column_names = [c.name for c in column_infos]
def _to_topk_tuples(sliced_table, categorical_features, weight_feature=None):
"""Generates tuples for computing top-k and uniques from input tables."""
slice_key, table = sliced_table
weight_column = table.column(weight_feature) if weight_feature else None
weight_array = weight_column.data.chunk(0) if weight_column else []
if weight_array:
flattened_weights = arrow_util.FlattenListArray(
weight_array).to_numpy()
for feature_column in table.columns:
feature_name = feature_column.name
# Skip the weight feature.
if feature_name == weight_feature:
continue
feature_path = types.FeaturePath([feature_name])
# if it's not a categorical feature nor a string feature, we don't bother
# with topk stats.
if not (feature_path in categorical_features
or feature_column.type.equals(pa.list_(pa.binary()))
or feature_column.type.equals(pa.list_(pa.string()))):
continue
value_array = feature_column.data.chunk(0)
flattened_values = arrow_util.FlattenListArray(value_array)
if weight_array and flattened_values:
if (pa.types.is_binary(flattened_values.type)
or pa.types.is_string(flattened_values.type)):
# no free conversion.
flattened_values_np = flattened_values.to_pandas()
else:
flattened_values_np = flattened_values.to_numpy()
indices = arrow_util.GetFlattenedArrayParentIndices(value_array)
weights_ndarray = flattened_weights[indices.to_numpy()]
for value, count, weight in _weighted_unique(
flattened_values_np, weights_ndarray):
yield (slice_key, feature_path.steps(), value), (count, weight)
else:
value_counts = arrow_util.ValueCounts(flattened_values)
values = value_counts.field('values').to_pylist()
counts = value_counts.field('counts').to_pylist()
for value, count in six.moves.zip(values, counts):
yield ((slice_key, feature_path.steps(), value), count)
def test_basic_stats_generator_empty_batch(self):
batches = [
pa.Table.from_arrays([pa.array([], type=pa.list_(pa.binary()))],
['a'])
]
expected_result = {
types.FeaturePath(['a']):
text_format.Parse(
"""
path {
step: 'a'
}
type: STRING
string_stats {
common_stats {
num_non_missing: 0
tot_num_values: 0
}
}
""", statistics_pb2.FeatureNameStatistics())
}
generator = basic_stats_generator.BasicStatsGenerator()
self.assertCombinerOutputEqual(batches, generator, expected_result)
value { float_list { value: [ 4.0 ] } }
}
feature {
key: "float_feature_2"
value { float_list { value: [ 5.0, 6.0 ] } }
}
feature {
key: "str_feature_1"
value { bytes_list { value: [ 'female' ] } }
}
feature {
key: "str_feature_2"
value { bytes_list { value: [ 'string', 'list' ] } }
}
}
''',
'decoded_table':
pa.Table.from_arrays([
pa.array([[0]], pa.list_(pa.int64())),
pa.array([[1, 2, 3]], pa.list_(pa.int64())),
pa.array([[4.0]], pa.list_(pa.float32())),
pa.array([[5.0, 6.0]], pa.list_(pa.float32())),
pa.array([[b'female']], pa.list_(pa.binary())),
pa.array([[b'string', b'list']], pa.list_(pa.binary()))
], [
'int_feature_1', 'int_feature_2', 'float_feature_1',
'float_feature_2', 'str_feature_1', 'str_feature_2'
])
},
]
def test_topk_uniques_combiner_with_single_bytes_feature(self):
# 'fa': 4 'a', 2 'b', 3 'c', 2 'd', 1 'e'
batches = [
pa.Table.from_arrays([
pa.array([['a', 'b', 'c', 'e'], ['a', 'c', 'd', 'a']],
type=pa.list_(pa.binary()))
], ['fa']),
pa.Table.from_arrays(
[pa.array([['a', 'b', 'c', 'd']], type=pa.list_(pa.binary()))],
['fa'])
]
# Note that if two feature values have the same frequency, the one with the
# lexicographically larger feature value will be higher in the order.
expected_result = {
types.FeaturePath(['fa']):
text_format.Parse(
"""
path {
step: 'fa'
}
type: STRING
string_stats {
unique: 5
top_values {
value: 'a'
frequency: 4
}
top_values {
value: 'c'
frequency: 3
}
top_values {
value: 'd'
frequency: 2
}
top_values {
value: 'b'
frequency: 2
}
rank_histogram {
buckets {
low_rank: 0
high_rank: 0
label: "a"
sample_count: 4.0
}
buckets {
low_rank: 1
high_rank: 1
label: "c"
sample_count: 3.0
}
buckets {
low_rank: 2
high_rank: 2
label: "d"
sample_count: 2.0
}
}
}""", statistics_pb2.FeatureNameStatistics())
}
generator = (top_k_uniques_combiner_stats_generator.
TopKUniquesCombinerStatsGenerator(
num_top_values=4, num_rank_histogram_buckets=3))
self.assertCombinerOutputEqual(batches, generator, expected_result)
def test_batch_serialized_examples(self):
examples = [
"""
features {
feature {
key: "a"
value { float_list { value: [ 1.0, 2.0 ] } }
}
feature {
key: "b"
value { bytes_list { value: [ 'a', 'b', 'c', 'e' ] } }
}
}""",
"""
features {
feature {
key: "a"
value { float_list { value: [ 3.0, 4.0, 5.0 ] } }
}
}""",
"""
features {
feature {
key: "b"
value { bytes_list { value: [ 'd', 'e', 'f' ] } }
}
feature {
key: "d"
value { int64_list { value: [ 10, 20, 30 ] } }
}
}""",
"""
features {
feature {
key: "b"
value { bytes_list { value: [ 'a', 'b', 'c' ] } }
}
}""",
"""
features {
feature {
key: "c"
value { bytes_list { value: [ 'd', 'e', 'f' ] } }
}
}""",
]
serialized_examples = [
text_format.Merge(example_pbtxt,
tf.train.Example()).SerializeToString()
for example_pbtxt in examples
]
expected_tables = [
pa.Table.from_arrays([
pa.array([[1.0, 2.0], [3.0, 4.0, 5.0]],
type=pa.list_(pa.float32())),
pa.array([['a', 'b', 'c', 'e'], None],
type=pa.list_(pa.binary()))
], ['a', 'b']),
pa.Table.from_arrays([
pa.array([['d', 'e', 'f'], ['a', 'b', 'c']],
type=pa.list_(pa.binary())),
pa.array([[10, 20, 30], None], type=pa.list_(pa.int64()))
], ['b', 'd']),
pa.Table.from_arrays(
[pa.array([['d', 'e', 'f']], type=pa.list_(pa.binary()))],
['c']),
]
with beam.Pipeline() as p:
result = (p
| beam.Create(serialized_examples)
| batch_util.BatchSerializedExamplesToArrowTables(
desired_batch_size=2))
util.assert_that(
result,
test_util.make_arrow_tables_equal_fn(self, expected_tables))
expected_output={
"int64_feature":
pa.array([[1, 2, 3], [4]], type=pa.list_(pa.int64())),
"uint64_feature":
pa.array([[1, 2, 3], None], type=pa.list_(pa.uint64())),
"int32_feature":
pa.array([[1, 2, 3], [4]], type=pa.list_(pa.int32())),
"uint32_feature":
pa.array([[1, 2, 3], None], type=pa.list_(pa.uint32())),
"float_feature":
pa.array([[1.], [2., 3., 4.]], type=pa.list_(pa.float32())),
"double_feature":
pa.array([[1.], [2., 3., 4.]], type=pa.list_(pa.float64())),
"bytes_feature":
pa.array([[b"abc", b"def"], [b"ghi"]],
type=pa.list_(pa.binary())),
"unicode_feature":
pa.array([[b"abc", b"def"], [b"ghi"]],
type=pa.list_(pa.string())),
}),
dict(testcase_name="mixed_unicode_and_bytes",
input_examples=[
{
"a": np.array([b"abc"], dtype=np.object),
},
{
"a": np.array([u"def"], dtype=np.object),
},
],
expected_output={
"a": pa.array([[b"abc"], [b"def"]], type=pa.list_(pa.binary()))
def test_stats_pipeline_with_examples_with_no_values(self):
tables = [
pa.Table.from_arrays([
pa.array([[]], type=pa.list_(pa.float32())),
pa.array([[]], type=pa.list_(pa.binary())),
pa.array([[]], type=pa.list_(pa.int32())),
pa.array([[2]]),
], ['a', 'b', 'c', 'w']),
pa.Table.from_arrays([
pa.array([[]], type=pa.list_(pa.float32())),
pa.array([[]], type=pa.list_(pa.binary())),
pa.array([[]], type=pa.list_(pa.int32())),
pa.array([[2]]),
], ['a', 'b', 'c', 'w']),
pa.Table.from_arrays([
pa.array([[]], type=pa.list_(pa.float32())),
pa.array([[]], type=pa.list_(pa.binary())),
pa.array([[]], type=pa.list_(pa.int32())),
pa.array([[2]]),
], ['a', 'b', 'c', 'w'])
]
expected_result = text_format.Parse(
"""
datasets{
num_examples: 3
features {
path {
step: 'a'
}
type: FLOAT
num_stats {
common_stats {
num_non_missing: 3
num_values_histogram {
buckets {
sample_count: 1.5
}
buckets {
sample_count: 1.5
}
type: QUANTILES
}
weighted_common_stats {
num_non_missing: 6
}
}
}
}
features {
path {
step: 'b'
}
type: STRING
string_stats {
common_stats {
num_non_missing: 3
num_values_histogram {
buckets {
sample_count: 1.5
}
buckets {
sample_count: 1.5
}
type: QUANTILES
}
weighted_common_stats {
num_non_missing: 6
}
}
}
}
features {
path {
step: 'c'
}
type: INT
num_stats {
common_stats {
num_non_missing: 3
num_values_histogram {
buckets {
sample_count: 1.5
}
buckets {
sample_count: 1.5
}
type: QUANTILES
}
weighted_common_stats {
num_non_missing: 6
}
}
}
}
}
""", statistics_pb2.DatasetFeatureStatisticsList())
with beam.Pipeline() as p:
options = stats_options.StatsOptions(
weight_feature='w',
num_top_values=1,
num_rank_histogram_buckets=1,
num_values_histogram_buckets=2,
num_histogram_buckets=1,
num_quantiles_histogram_buckets=1,
epsilon=0.001)
result = (p | beam.Create(tables)
| stats_api.GenerateStatistics(options))
util.assert_that(
result,
test_util.make_dataset_feature_stats_list_proto_equal_fn(
self, expected_result))
