def get_expected(data, col_properties):
expected = []
_map_col_types = {'INT': 'int_col', 'DOUBLE': 'real_col', 'STR': 'str_col'}
_map_col_types.update(
{k: 'str_col'
for k in _pandas_loaders.GEO_TYPE_NAMES})
isnull = data.isnull()
for prop in col_properties:
nulls = isnull[prop['name']].tolist()
if prop['is_array']:
arr_col = []
for v in data[prop['name']]:
arr_col.append(
TColumn(data=TColumnData(
**{_map_col_types[prop['type']]: v})), )
col = TColumn(data=TColumnData(arr_col=arr_col), nulls=nulls)
elif prop['type'] in _pandas_loaders.GEO_TYPE_NAMES:
col = TColumn(
data=TColumnData(
**{
_map_col_types[prop['type']]:
data[prop['name']].apply(lambda g: g.wkt)
}),
nulls=nulls,
)
else:
col = TColumn(
data=TColumnData(
**{_map_col_types[prop['type']]: data[prop['name']]}),
nulls=nulls,
)
expected.append(col)
return expected
def get_expected(data, col_properties):
expected = []
_map_col_types = {
'INT': 'int_col',
'DOUBLE': 'real_col',
'STR': 'str_col',
'TIMESTAMP': 'int_col',
'DECIMAL': 'int_col',
}
_map_col_types.update(
{k: 'str_col'
for k in _pandas_loaders.GEO_TYPE_NAMES})
isnull = data.isnull()
for prop in col_properties:
nulls = isnull[prop['name']].tolist()
if prop['is_array']:
arr_col = []
for v in data[prop['name']]:
arr_col.append(
TColumn(data=TColumnData(
**{_map_col_types[prop['type']]: v})), )
col = TColumn(data=TColumnData(arr_col=arr_col), nulls=nulls)
elif prop['type'] in _pandas_loaders.GEO_TYPE_NAMES:
col = TColumn(
data=TColumnData(
**{
_map_col_types[prop['type']]:
data[prop['name']].apply(lambda g: g.wkt)
}),
nulls=nulls,
)
else:
if prop['type'] == 'TIMESTAMP':
# convert datetime to epoch
if data[prop['name']].dt.nanosecond.sum():
data[prop['name']] = data[prop['name']].astype(int)
else:
data[prop['name']] = (data[prop['name']].astype(int) //
10**9)
elif prop['type'] == 'DECIMAL':
# data = (data * 10 ** precision).astype(int) \
# * 10 ** (scale - precision)
data[prop['name']] = (data[prop['name']] * 10**
prop['precision']).astype(int) * 10**(
prop['scale'] - prop['precision'])
col = TColumn(
data=TColumnData(
**{_map_col_types[prop['type']]: data[prop['name']]}),
nulls=nulls,
)
expected.append(col)
return expected
def test_build_table_columnar_nulls(self):
common_col_params = dict(
nullable=True,
scale=0,
comp_param=0,
encoding='NONE',
is_array=False,
)
col_types = [
ColumnDetails(name='boolean_',
type='BOOL',
precision=0,
**common_col_params),
ColumnDetails(name='int_',
type='INT',
precision=0,
**common_col_params),
ColumnDetails(name='bigint_',
type='BIGINT',
precision=0,
**common_col_params),
ColumnDetails(name='double_',
type='DOUBLE',
precision=0,
**common_col_params),
ColumnDetails(name='varchar_',
type='STR',
precision=0,
**common_col_params),
ColumnDetails(name='text_',
type='STR',
precision=0,
**common_col_params),
ColumnDetails(name='time_',
type='TIME',
precision=0,
**common_col_params),
ColumnDetails(
name='timestamp_',
type='TIMESTAMP',
**common_col_params,
precision=0,
),
ColumnDetails(name='date_',
type='DATE',
precision=0,
**common_col_params),
]
data = pd.DataFrame({
'boolean_': [True, False, None],
# Currently Pandas does not support storing None or NaN
# in integer columns, so int cols with null
# need to be objects. This means our type detection will be
# unreliable since if there is no number outside the int32
# bounds in a column with nulls then we will be assuming int
'int_':
np.array([0, 1, None], dtype=np.object),
'bigint_':
np.array([0, 9223372036854775807, None], dtype=np.object),
'double_':
np.array([0, 1, None], dtype=np.float64),
'varchar_': ['a', 'b', None],
'text_': ['a', 'b', None],
'time_': [datetime.time(0, 11, 59),
datetime.time(13), None],
'timestamp_': [
pd.Timestamp('2016'),
pd.Timestamp('2017'),
None,
],
'date_': [
datetime.date(1001, 1, 1),
datetime.date(2017, 1, 1),
None,
],
})
result = _pandas_loaders.build_input_columnar(
data,
preserve_index=False,
col_names=data.columns,
col_types=col_types,
)
nulls = [False, False, True]
bool_na = -128
int_na = -2147483648
bigint_na = -9223372036854775808
ns_na = -9223372037
double_na = 0
expected = [
TColumn(TColumnData(int_col=[1, 0, bool_na]), nulls=nulls),
TColumn(
TColumnData(int_col=np.array([0, 1, int_na], dtype=np.int32)),
nulls=nulls,
), # noqa
TColumn(
TColumnData(int_col=np.array(
[0, 9223372036854775807, bigint_na], dtype=np.int64)),
nulls=nulls,
), # noqa
TColumn(
TColumnData(
real_col=np.array([0, 1, double_na], dtype=np.float64)),
nulls=nulls,
), # noqa
TColumn(TColumnData(str_col=['a', 'b', '']), nulls=nulls),
TColumn(TColumnData(str_col=['a', 'b', '']), nulls=nulls),
TColumn(TColumnData(int_col=[719, 46800, bigint_na]), nulls=nulls),
TColumn(
TColumnData(int_col=[1451606400, 1483228800, ns_na]),
nulls=nulls,
), # noqa
TColumn(
TColumnData(int_col=[-30578688000, 1483228800, bigint_na]),
nulls=nulls,
), # noqa
]
assert_columnar_equal(result[0], expected)
def test_build_table_columnar_pandas(self):
common_col_params = dict(
nullable=True,
precision=0,
scale=0,
comp_param=0,
encoding='NONE',
is_array=False,
)
col_types = [
ColumnDetails(name='boolean_', type='BOOL', **common_col_params),
ColumnDetails(name='smallint_',
type='SMALLINT',
**common_col_params),
ColumnDetails(name='int_', type='INT', **common_col_params),
ColumnDetails(name='bigint_', type='BIGINT', **common_col_params),
ColumnDetails(name='float_', type='FLOAT', **common_col_params),
ColumnDetails(name='double_', type='DOUBLE', **common_col_params),
ColumnDetails(name='varchar_', type='STR', **common_col_params),
ColumnDetails(name='text_', type='STR', **common_col_params),
ColumnDetails(name='time_', type='TIME', **common_col_params),
ColumnDetails(
name='timestamp_',
type='TIMESTAMP',
nullable=True,
precision=0,
scale=0,
comp_param=0,
encoding='NONE',
is_array=False,
),
ColumnDetails(name='date_', type='DATE', **common_col_params),
]
data = pd.DataFrame({
'boolean_': [True, False],
'smallint_':
np.array([0, 1], dtype=np.int16),
'int_':
np.array([0, 1], dtype=np.int32),
'bigint_':
np.array([0, 1], dtype=np.int64),
'float_':
np.array([0, 1], dtype=np.float32),
'double_':
np.array([0, 1], dtype=np.float64),
'varchar_': ['a', 'b'],
'text_': ['a', 'b'],
'time_': [datetime.time(0, 11, 59),
datetime.time(13)],
'timestamp_': [pd.Timestamp('2016'),
pd.Timestamp('2017')],
'date_': [
datetime.date(2016, 1, 1),
datetime.date(2017, 1, 1),
],
})
result = _pandas_loaders.build_input_columnar(
data,
preserve_index=False,
col_names=data.columns,
col_types=col_types,
)
nulls = [False, False]
expected = [
TColumn(TColumnData(int_col=[True, False]), nulls=nulls),
TColumn(
TColumnData(int_col=np.array([0, 1], dtype=np.int16)),
nulls=nulls,
), # noqa
TColumn(
TColumnData(int_col=np.array([0, 1], dtype=np.int32)),
nulls=nulls,
), # noqa
TColumn(
TColumnData(int_col=np.array([0, 1], dtype=np.int64)),
nulls=nulls,
), # noqa
TColumn(
TColumnData(real_col=np.array([0, 1], dtype=np.float32)),
nulls=nulls,
), # noqa
TColumn(
TColumnData(real_col=np.array([0, 1], dtype=np.float64)),
nulls=nulls,
), # noqa
TColumn(TColumnData(str_col=['a', 'b']), nulls=nulls),
TColumn(TColumnData(str_col=['a', 'b']), nulls=nulls),
TColumn(TColumnData(int_col=[719, 46800]), nulls=nulls),
TColumn(TColumnData(int_col=[1451606400, 1483228800]),
nulls=nulls), # noqa
TColumn(TColumnData(int_col=[1451606400, 1483228800]),
nulls=nulls),
]
assert_columnar_equal(result[0], expected)