def testProjectPrune(self):
expr = self.expr.select('name', 'id')
new_expr = ColumnPruning(expr.to_dag()).prune()
self.assertIsInstance(new_expr, ProjectCollectionExpr)
self.assertIsNotNone(new_expr.input._source_data)
expected = 'SELECT t1.`name`, t1.`id` \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1'
self.assertEqual(expected,
ODPSEngine(self.odps).compile(expr, prettify=False))
expr = self.expr[Scalar(3).rename('const'),
NullScalar('string').rename('string_const'),
self.expr.id]
expected = 'SELECT 3 AS `const`, CAST(NULL AS STRING) AS `string_const`, t1.`id` \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.select(
pt=BuiltinFunction('max_pt', args=(self.expr._source_data.name, )))
expected = "SELECT max_pt('pyodps_test_expr_table') AS `pt` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def testFilterPrune(self):
expr = self.expr.filter(self.expr.name == 'name1')
expr = expr['name', 'id']
new_expr = ColumnPruning(expr.to_dag()).prune()
self.assertIsInstance(new_expr.input, FilterCollectionExpr)
self.assertNotIsInstance(new_expr.input.input, ProjectCollectionExpr)
self.assertIsNotNone(new_expr.input.input._source_data)
expected = 'SELECT t1.`name`, t1.`id` \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
'WHERE t1.`name` == \'name1\''
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.filter(self.expr.name == 'name1')
new_expr = ColumnPruning(expr.to_dag()).prune()
self.assertIsInstance(new_expr, FilterCollectionExpr)
self.assertIsNotNone(new_expr.input._source_data)
expr = self.expr.filter(self.expr.id.isin(self.expr3.id))
expected = 'SELECT * \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
'WHERE t1.`id` IN (SELECT t3.`id` FROM ( ' \
'SELECT t2.`id` FROM mocked_project.`pyodps_test_expr_table2` t2 ) t3)'
self.assertTrue(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def testUnionPrune(self):
left = self.expr.select('name', 'id')
right = self.expr3.select(
self.expr3.fid.astype('int').rename('id'), self.expr3.name)
expr = left.union(right)['id']
expected = "SELECT t3.`id` \n" \
"FROM (\n" \
" SELECT t1.`id` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
" UNION ALL\n" \
" SELECT CAST(t2.`fid` AS BIGINT) AS `id` \n" \
" FROM mocked_project.`pyodps_test_expr_table2` t2\n" \
") t3"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.union(self.expr2)
expected = 'SELECT * \n' \
'FROM (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
' UNION ALL\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table2` t2\n' \
') t3'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid'],
datatypes('string', 'int64', 'float64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
self.engine = ODPSEngine(self.odps)
def testFilterPushdownThroughProjection(self):
expr = self.expr[self.expr.id + 1, 'name'][lambda x: x.id < 10]
expected = 'SELECT t1.`id` + 1 AS `id`, t1.`name` \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
'WHERE (t1.`id` + 1) < 10'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr['name', self.expr.id ** 2]\
.filter(lambda x: x.name == 'name1').filter(lambda x: x.id < 3)
expected = "SELECT t1.`name`, CAST(POW(t1.`id`, 2) AS BIGINT) AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE (t1.`name` == 'name1') AND ((CAST(POW(t1.`id`, 2) AS BIGINT)) < 3)"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr['name', self.expr.id + 1].filter(
lambda x: x.name == 'name1')[lambda x: 'tt' + x.name,
'id'].filter(lambda x: x.id < 3)
expected = "SELECT CONCAT('tt', t1.`name`) AS `name`, t1.`id` + 1 AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE (t1.`name` == 'name1') AND ((t1.`id` + 1) < 3)"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.filter(self.expr.name == 'name1').select(
'name', lambda x: (x.id + 1) * 2)[lambda x: 'tt' + x.name,
'id'].filter(lambda x: x.id < 3)
expected = "SELECT CONCAT('tt', t1.`name`) AS `name`, (t1.`id` + 1) * 2 AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE (((t1.`id` + 1) * 2) < 3) AND (t1.`name` == 'name1')"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.filter(
self.expr.id.between(2, 6),
self.expr.name.lower().contains('pyodps',
regex=False)).name.nunique()
expected = "SELECT COUNT(DISTINCT t2.`name`) AS `name_nunique` \n" \
"FROM (\n" \
" SELECT t1.`name`, t1.`id` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
" WHERE ((t1.`id` >= 2) AND (t1.`id` <= 6)) AND INSTR(TOLOWER(t1.`name`), 'pyodps') > 0 \n" \
") t2"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal', 'datetime'))
self.schema = df_schema_to_odps_schema(schema)
table_name = 'pyodps_test_engine_table'
self.odps.delete_table(table_name, if_exists=True)
self.table = self.odps.create_table(
name='pyodps_test_engine_table', schema=self.schema)
self.expr = CollectionExpr(_source_data=self.table, _schema=schema)
self.engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def testSamplePrune(self):
expr = self.expr['name', 'id'].sample(parts=5)['id', ]
expected = "SELECT t1.`id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE SAMPLE(5, 1)"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid'],
datatypes('string', 'int64', 'float64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
self.engine = ODPSEngine(self.odps)
def testDistinctPrune(self):
expr = self.expr.distinct(self.expr.id + 1, self.expr.name)['name', ]
expected = "SELECT t2.`name` \n" \
"FROM (\n" \
" SELECT DISTINCT t1.`id` + 1 AS `id`, t1.`name` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
") t2"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(
['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal',
'datetime'))
self.schema = df_schema_to_odps_schema(schema)
import pandas as pd
self.df = pd.DataFrame(None, columns=schema.names)
self.expr = CollectionExpr(_source_data=self.df, _schema=schema)
self.engine = PandasEngine(self.odps)
self.odps_engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def testFilterPushdownThroughProjection(self):
expr = self.expr[self.expr.id + 1, 'name'][lambda x: x.id < 10]
expected = 'SELECT t1.`id` + 1 AS `id`, t1.`name` \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
'WHERE (t1.`id` + 1) < 10'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr['name', self.expr.id ** 2]\
.filter(lambda x: x.name == 'name1').filter(lambda x: x.id < 3)
expected = "SELECT t1.`name`, CAST(POW(t1.`id`, 2) AS BIGINT) AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE (t1.`name` == 'name1') AND ((CAST(POW(t1.`id`, 2) AS BIGINT)) < 3)"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr['name', self.expr.id + 1].filter(
lambda x: x.name == 'name1')[lambda x: 'tt' + x.name,
'id'].filter(lambda x: x.id < 3)
expected = "SELECT CONCAT('tt', t1.`name`) AS `name`, t1.`id` + 1 AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE (t1.`name` == 'name1') AND ((t1.`id` + 1) < 3)"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.filter(self.expr.name == 'name1').select(
'name', lambda x: (x.id + 1) * 2)[lambda x: 'tt' + x.name,
'id'].filter(lambda x: x.id < 3)
expected = "SELECT CONCAT('tt', t1.`name`) AS `name`, (t1.`id` + 1) * 2 AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE (((t1.`id` + 1) * 2) < 3) AND (t1.`name` == 'name1')"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def testGroupbyPrune(self):
expr = self.expr.groupby('name').agg(id=self.expr.id.max())
expr = expr[expr.id < 0]['name', ]
expected = "SELECT t1.`name` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"GROUP BY t1.`name` \n" \
"HAVING MAX(t1.`id`) < 0"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.groupby('name').agg(id=self.expr.id.max())
expr = expr[expr.id < 0]['id', ]
expected = "SELECT MAX(t1.`id`) AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"GROUP BY t1.`name` \n" \
"HAVING MAX(t1.`id`) < 0"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def testSlicePrune(self):
expr = self.expr.filter(self.expr.fid < 0)[:4]['name',
lambda x: x.id + 1]
new_expr = ColumnPruning(expr.to_dag()).prune()
self.assertIsNotNone(new_expr.input.input.input._source_data)
expected = "SELECT t1.`name`, t1.`id` + 1 AS `id` \n" \
"FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
"WHERE t1.`fid` < 0 \n" \
"LIMIT 4"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def testJoinPrune(self):
left = self.expr.select(self.expr, type='normal')
right = self.expr3[:4]
joined = left.left_join(right, on='id')
expr = joined.id_x.rename('id')
expected = "SELECT t2.`id` \n" \
"FROM (\n" \
" SELECT t1.`id` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1\n" \
") t2 \n" \
"LEFT OUTER JOIN \n" \
" (\n" \
" SELECT t3.`id` \n" \
" FROM mocked_project.`pyodps_test_expr_table2` t3 \n" \
" LIMIT 4\n" \
" ) t4\n" \
"ON t2.`id` == t4.`id`"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
joined = self.expr.join(self.expr2, 'name')
expected = 'SELECT t1.`name`, t1.`id` AS `id_x`, t1.`fid` AS `fid_x`, ' \
't1.`isMale` AS `isMale_x`, t1.`scale` AS `scale_x`, ' \
't1.`birth` AS `birth_x`, t1.`ds` AS `ds_x`, t2.`id` AS `id_y`, ' \
't2.`fid` AS `fid_y`, t2.`isMale` AS `isMale_y`, t2.`scale` AS `scale_y`, ' \
't2.`birth` AS `birth_y`, t2.`ds` AS `ds_y` \n' \
'FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
'INNER JOIN \n' \
' mocked_project.`pyodps_test_expr_table2` t2\n' \
'ON t1.`name` == t2.`name`'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(joined, prettify=False)))
def testSortPrune(self):
expr = self.expr[self.expr.exclude('name'),
self.expr.name.rename('name2')].sort('name2')['id',
'fid']
expected = "SELECT t2.`id`, t2.`fid` \n" \
"FROM (\n" \
" SELECT t1.`id`, t1.`fid`, t1.`name` AS `name2` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
" ORDER BY name2 \n" \
" LIMIT 10000\n" \
") t2"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal', 'datetime'))
self.schema = df_schema_to_odps_schema(schema)
import pandas as pd
self.df = pd.DataFrame(None, columns=schema.names)
self.expr = CollectionExpr(_source_data=self.df, _schema=schema)
self.engine = PandasEngine(self.odps)
self.odps_engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def testFilterPushdownThroughUnion(self):
expr = self.expr['name', 'id'].union(self.expr2['id', 'name'])
expr = expr.filter(expr.id + 1 < 3)
expected = 'SELECT * \n' \
'FROM (\n' \
' SELECT t1.`name`, t1.`id` \n' \
' FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
' WHERE (t1.`id` + 1) < 3 \n' \
' UNION ALL\n' \
' SELECT t2.`name`, t2.`id` \n' \
' FROM mocked_project.`pyodps_test_expr_table2` t2 \n' \
' WHERE (t2.`id` + 1) < 3\n' \
') t3'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal', 'datetime'))
self.schema = df_schema_to_odps_schema(schema)
table_name = 'pyodps_test_engine_table'
self.odps.delete_table(table_name, if_exists=True)
self.table = self.odps.create_table(
name='pyodps_test_engine_table', schema=self.schema)
self.expr = CollectionExpr(_source_data=self.table, _schema=schema)
self.engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def testFilterPartitionPrune(self):
expr = self.expr.filter_partition('ds=today')[lambda x: x.fid < 0][
'name', lambda x: x.id + 1]
new_expr = ColumnPruning(expr.to_dag()).prune()
self.assertEqual(set(new_expr.input.input.schema.names),
set(['name', 'id', 'fid']))
expected = "SELECT t2.`name`, t2.`id` + 1 AS `id` \n" \
"FROM (\n" \
" SELECT t1.`name`, t1.`id`, t1.`fid` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
" WHERE t1.`ds` == 'today' \n" \
") t2 \n" \
"WHERE t2.`fid` < 0"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
def testMutatePrune(self):
expr = self.expr[self.expr.exclude('birth'),
self.expr.fid.astype('int').rename('new_id')]
expr = expr[expr,
expr.groupby('name').
mutate(lambda x: x.new_id.cumsum().rename('new_id_sum'))]
expr = expr[expr.new_id, expr.new_id_sum]
expected = "SELECT t2.`new_id`, t2.`new_id_sum` \n" \
"FROM (\n" \
" SELECT CAST(t1.`fid` AS BIGINT) AS `new_id`, " \
"SUM(CAST(t1.`fid` AS BIGINT)) OVER (PARTITION BY t1.`name`) AS `new_id_sum` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
") t2"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal', 'datetime'))
self.schema = df_schema_to_odps_schema(schema)
table_name = 'pyodps_test_engine_table'
self.odps.delete_table(table_name, if_exists=True)
self.table = self.odps.create_table(
name='pyodps_test_engine_table', schema=self.schema)
self.expr = CollectionExpr(_source_data=self.table, _schema=schema)
self.engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def _gen_data(self, rows=None, data=None, nullable_field=None, value_range=None):
if data is None:
data = []
for _ in range(rows):
record = []
for t in self.schema.types:
method = getattr(self, '_gen_random_%s' % t.name)
if t.name == 'bigint':
record.append(method(value_range=value_range))
else:
record.append(method())
data.append(record)
if nullable_field is not None:
j = self.schema._name_indexes[nullable_field]
for i, l in enumerate(data):
if i % 2 == 0:
data[i][j] = None
self.odps.write_table(self.table, 0, [self.table.new_record(values=d) for d in data])
return data
def testTunnelCases(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr.count()
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(10, result)
expr = self.expr.name.count()
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(10, result)
res = self.engine._handle_cases(self.expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(data, result)
expr = self.expr['name', self.expr.id.rename('new_id')]
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual([it[:2] for it in data], result)
table_name = 'pyodps_test_engine_partitioned'
self.odps.delete_table(table_name, if_exists=True)
df = self.engine.persist(self.expr, table_name, partitions=['name'])
try:
expr = df.count()
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertIsNone(res)
expr = df[df.name == data[0][0]]['fid', 'id'].count()
expr = self.engine._pre_process(expr)
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(res, 0)
expr = df[df.name == data[0][0]]['fid', 'id']
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(len(res), 0)
finally:
self.odps.delete_table(table_name, if_exists=True)
df = self.engine.persist(self.expr, table_name, partitions=['name', 'id'])
try:
expr = df.count()
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertIsNone(res)
expr = df[(df.name == data[0][0]) & (df.id == data[0][1])]['fid', 'ismale'].count()
expr = self.engine._pre_process(expr)
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(res, 0)
expr = df[(df.name == data[0][0]) & (df.id == data[0][1])]['fid', 'ismale']
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(len(res), 0)
finally:
self.odps.delete_table(table_name, if_exists=True)
def testAsync(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr.id.sum()
res = self.engine.execute(expr, async=True)
self.assertNotEqual(res.instance.status, Instance.Status.TERMINATED)
res.wait()
self.assertEqual(sum(it[1] for it in data), res.fetch())
def testBase(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr[self.expr.id < 10]['name', lambda x: x.id]
result = self._get_result(self.engine.execute(expr).values)
self.assertEqual(len([it for it in data if it[1] < 10]), len(result))
if len(result) > 0:
self.assertEqual(2, len(result[0]))
expr = self.expr[Scalar(3).rename('const'), self.expr.id, (self.expr.id + 1).rename('id2')]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual([c.name for c in res.columns], ['const', 'id', 'id2'])
self.assertTrue(all(it[0] == 3 for it in result))
self.assertEqual(len(data), len(result))
self.assertEqual([it[1]+1 for it in data], [it[2] for it in result])
expr = self.expr.sort('id')[:5]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual(sorted(data, key=lambda it: it[1])[:5], result)
expr = self.expr.sort('id')[:5]
# test do not use tunnel
res = self.engine.execute(expr, use_tunnel=False)
result = self._get_result(res.values)
self.assertEqual(sorted(data, key=lambda it: it[1])[:5], result)
def testElement(self):
data = self._gen_data(5, nullable_field='name')
fields = [
self.expr.name.isnull().rename('name1'),
self.expr.name.notnull().rename('name2'),
self.expr.name.fillna('test').rename('name3'),
self.expr.id.isin([1, 2, 3]).rename('id1'),
self.expr.id.isin(self.expr.fid.astype('int')).rename('id2'),
self.expr.id.notin([1, 2, 3]).rename('id3'),
self.expr.id.notin(self.expr.fid.astype('int')).rename('id4'),
self.expr.id.between(self.expr.fid, 3).rename('id5'),
self.expr.name.fillna('test').switch('test', 'test' + self.expr.name.fillna('test'),
'test2', 'test2' + self.expr.name.fillna('test'),
default=self.expr.name).rename('name4'),
self.expr.id.cut([100, 200, 300],
labels=['xsmall', 'small', 'large', 'xlarge'],
include_under=True, include_over=True).rename('id6')
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual(len([it for it in data if it[0] is None]),
len([it[0] for it in result if it[0]]))
self.assertEqual(len([it[0] for it in data if it[0] is not None]),
len([it[1] for it in result if it[1]]))
self.assertEqual([(it[0] if it[0] is not None else 'test') for it in data],
[it[2] for it in result])
self.assertEqual([(it[1] in (1, 2, 3)) for it in data],
[it[3] for it in result])
fids = [int(it[2]) for it in data]
self.assertEqual([(it[1] in fids) for it in data],
[it[4] for it in result])
self.assertEqual([(it[1] not in (1, 2, 3)) for it in data],
[it[5] for it in result])
self.assertEqual([(it[1] not in fids) for it in data],
[it[6] for it in result])
self.assertEqual([(it[2] <= it[1] <= 3) for it in data],
[it[7] for it in result])
self.assertEqual([to_str('testtest' if it[0] is None else it[0]) for it in data],
[to_str(it[8]) for it in result])
def get_val(val):
if val <= 100:
return 'xsmall'
elif 100 < val <= 200:
return 'small'
elif 200 < val <= 300:
return 'large'
else:
return 'xlarge'
self.assertEqual([to_str(get_val(it[1])) for it in data], [to_str(it[9]) for it in result])
def testArithmetic(self):
data = self._gen_data(5, value_range=(-1000, 1000))
fields = [
(self.expr.id + 1).rename('id1'),
(self.expr.fid - 1).rename('fid1'),
(self.expr.scale * 2).rename('scale1'),
(self.expr.scale + self.expr.id).rename('scale2'),
(self.expr.id / 2).rename('id2'),
(self.expr.id ** -2).rename('id3'),
abs(self.expr.id).rename('id4'),
(~self.expr.id).rename('id5'),
(-self.expr.fid).rename('fid2'),
(~self.expr.isMale).rename('isMale1'),
(-self.expr.isMale).rename('isMale2'),
(self.expr.id // 2).rename('id6'),
(self.expr.birth + day(1).rename('birth1')),
(self.expr.birth - (self.expr.birth - millisecond(10))).rename('birth2'),
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual([it[1] + 1 for it in data],
[it[0] for it in result])
self.assertAlmostEqual([it[2] - 1 for it in data],
[it[1] for it in result])
self.assertEqual([it[4] * 2 for it in data],
[it[2] for it in result])
self.assertEqual([it[4] + it[1] for it in data],
[it[3] for it in result])
self.assertAlmostEqual([float(it[1]) / 2 for it in data],
[it[4] for it in result])
self.assertEqual([int(it[1] ** -2) for it in data],
[it[5] for it in result])
self.assertEqual([abs(it[1]) for it in data],
[it[6] for it in result])
self.assertEqual([~it[1] for it in data],
[it[7] for it in result])
self.assertAlmostEqual([-it[2] for it in data],
[it[8] for it in result])
self.assertEqual([not it[3] for it in data],
[it[9] for it in result])
self.assertEqual([it[1] // 2 for it in data],
[it[11] for it in result])
self.assertEqual([it[5] + timedelta(days=1) for it in data],
[it[12] for it in result])
self.assertEqual([10] * len(data), [it[13] for it in result])
def testMath(self):
data = self._gen_data(5, value_range=(1, 90))
import numpy as np
methods_to_fields = [
(np.sin, self.expr.id.sin()),
(np.cos, self.expr.id.cos()),
(np.tan, self.expr.id.tan()),
(np.sinh, self.expr.id.sinh()),
(np.cosh, self.expr.id.cosh()),
(np.tanh, self.expr.id.tanh()),
(np.log, self.expr.id.log()),
(np.log2, self.expr.id.log2()),
(np.log10, self.expr.id.log10()),
(np.log1p, self.expr.id.log1p()),
(np.exp, self.expr.id.exp()),
(np.expm1, self.expr.id.expm1()),
(np.arccosh, self.expr.id.arccosh()),
(np.arcsinh, self.expr.id.arcsinh()),
(np.arctanh, self.expr.id.arctanh()),
(np.arctan, self.expr.id.arctan()),
(np.sqrt, self.expr.id.sqrt()),
(np.abs, self.expr.id.abs()),
(np.ceil, self.expr.id.ceil()),
(np.floor, self.expr.id.floor()),
(np.trunc, self.expr.id.trunc()),
]
fields = [it[1].rename('id'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[1]) for it in data]
second = [it[i] for it in result]
self.assertEqual(len(first), len(second))
for it1, it2 in zip(first, second):
if np.isnan(it1) and np.isnan(it2):
continue
self.assertAlmostEqual(it1, it2)
def testString(self):
data = self._gen_data(5)
methods_to_fields = [
(lambda s: s.capitalize(), self.expr.name.capitalize()),
(lambda s: data[0][0] in s, self.expr.name.contains(data[0][0], regex=False)),
(lambda s: s.count(data[0][0]), self.expr.name.count(data[0][0])),
(lambda s: s.endswith(data[0][0]), self.expr.name.endswith(data[0][0])),
(lambda s: s.startswith(data[0][0]), self.expr.name.startswith(data[0][0])),
(lambda s: s.find(data[0][0]), self.expr.name.find(data[0][0])),
(lambda s: s.rfind(data[0][0]), self.expr.name.rfind(data[0][0])),
(lambda s: s.replace(data[0][0], 'test'), self.expr.name.replace(data[0][0], 'test')),
(lambda s: s[0], self.expr.name.get(0)),
(lambda s: len(s), self.expr.name.len()),
(lambda s: s.ljust(10), self.expr.name.ljust(10)),
(lambda s: s.ljust(20, '*'), self.expr.name.ljust(20, fillchar='*')),
(lambda s: s.rjust(10), self.expr.name.rjust(10)),
(lambda s: s.rjust(20, '*'), self.expr.name.rjust(20, fillchar='*')),
(lambda s: s * 4, self.expr.name.repeat(4)),
(lambda s: s[2: 10: 2], self.expr.name.slice(2, 10, 2)),
(lambda s: s[-5: -1], self.expr.name.slice(-5, -1)),
(lambda s: s.title(), self.expr.name.title()),
(lambda s: s.rjust(20, '0'), self.expr.name.zfill(20)),
(lambda s: s.isalnum(), self.expr.name.isalnum()),
(lambda s: s.isalpha(), self.expr.name.isalpha()),
(lambda s: s.isdigit(), self.expr.name.isdigit()),
(lambda s: s.isspace(), self.expr.name.isspace()),
(lambda s: s.isupper(), self.expr.name.isupper()),
(lambda s: s.istitle(), self.expr.name.istitle()),
(lambda s: to_str(s).isnumeric(), self.expr.name.isnumeric()),
(lambda s: to_str(s).isdecimal(), self.expr.name.isdecimal()),
]
fields = [it[1].rename('id'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[0]) for it in data]
second = [it[i] for it in result]
self.assertEqual(first, second)
def testApply(self):
data = self._gen_data(5)
def my_func(row):
return row.name,
expr = self.expr['name', 'id'].apply(my_func, axis=1, names='name')
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([r[0] for r in result], [r[0] for r in data])
def my_func2(row):
yield len(row.name)
yield row.id
expr = self.expr['name', 'id'].apply(my_func2, axis=1, names='cnt', types='int')
res = self.engine.execute(expr)
result = self._get_result(res)
def gen_expected(data):
for r in data:
yield len(r[0])
yield r[1]
self.assertEqual([r[0] for r in result], [r for r in gen_expected(data)])
def testDatetime(self):
data = self._gen_data(5)
import pandas as pd
methods_to_fields = [
(lambda s: list(s.birth.dt.year.values), self.expr.birth.year),
(lambda s: list(s.birth.dt.month.values), self.expr.birth.month),
(lambda s: list(s.birth.dt.day.values), self.expr.birth.day),
(lambda s: list(s.birth.dt.hour.values), self.expr.birth.hour),
(lambda s: list(s.birth.dt.minute.values), self.expr.birth.minute),
(lambda s: list(s.birth.dt.second.values), self.expr.birth.second),
(lambda s: list(s.birth.dt.weekofyear.values), self.expr.birth.weekofyear),
(lambda s: list(s.birth.dt.dayofweek.values), self.expr.birth.dayofweek),
(lambda s: list(s.birth.dt.weekday.values), self.expr.birth.weekday),
(lambda s: list(s.birth.dt.date.values), self.expr.birth.date),
(lambda s: list(s.birth.dt.strftime('%Y%d')), self.expr.birth.strftime('%Y%d'))
]
fields = [it[1].rename('birth'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
def conv(v):
if isinstance(v, pd.Timestamp):
return v.to_datetime().date()
else:
return v
second = [conv(it[i]) for it in result]
self.assertEqual(first, second)
def testSortDistinct(self):
data = [
['name1', 4, None, None, None, None],
['name2', 2, None, None, None, None],
['name1', 4, None, None, None, None],
['name1', 3, None, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.sort(['name', -self.expr.id]).distinct(['name', lambda x: x.id + 1])[:50]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 3)
expected = [
['name1', 5],
['name1', 4],
['name2', 3]
]
self.assertEqual(expected, result)
def testGroupbyAggregation(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(['name', 'id'])[lambda x: x.fid.min() * 2 < 8] \
.agg(self.expr.fid.max() + 1, new_id=self.expr.id.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 5.1, 6],
['name2', 2, 4.5, 2]
]
result = sorted(result, key=lambda k: k[0])
self.assertEqual(expected, result)
field = self.expr.groupby('name').sort(['id', -self.expr.fid]).row_number()
expr = self.expr['name', 'id', 'fid', field]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 4.1, 1],
['name1', 3, 2.2, 2],
['name1', 4, 5.3, 3],
['name1', 4, 4.2, 4],
['name2', 2, 3.5, 1],
]
result = sorted(result, key=lambda k: (k[0], k[1], -k[2]))
self.assertEqual(expected, result)
expr = self.expr.name.value_counts()[:25]
expected = [
['name1', 4],
['name2', 1]
]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.name.topk(25)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.groupby('name').count()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([it[1:] for it in expected], result)
expected = [
['name1', 2],
['name2', 1]
]
expr = self.expr.groupby('name').id.nunique()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([it[1:] for it in expected], result)
expr = self.expr[self.expr['id'] > 2].name.value_counts()[:25]
expected = [
['name1', 4]
]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
def testJoinGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2, _schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [
['name1', 4, -1],
['name2', 1, -2]
]
self.odps.write_table(table2, 0, [table2.new_record(values=d) for d in data2])
expr = self.expr.join(expr2, on='name')[self.expr]
expr = expr.groupby('id').agg(expr.fid.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
import pandas as pd
expected = pd.DataFrame(data, columns=self.expr.schema.names).groupby('id').agg({'fid': 'sum'})\
.reset_index().values.tolist()
for it in zip(sorted(expected, key=lambda it: it[0]), sorted(result, key=lambda it: it[0])):
self.assertAlmostEqual(it[0][0], it[1][0])
self.assertAlmostEqual(it[0][1], it[1][1])
def testFilterGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(['name']).agg(id=self.expr.id.max())[lambda x: x.id > 3]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 1)
expected = [
['name1', 4]
]
self.assertEqual(expected, result)
def testWindowRewrite(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr[self.expr.id - self.expr.id.mean() < 10][
[lambda x: x.id - x.id.max()]][[lambda x: x.id - x.id.min()]][lambda x: x.id - x.id.std() > 0]
res = self.engine.execute(expr)
result = self._get_result(res)
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
expected = df.id - df.id.max()
expected = expected - expected.min()
expected = list(expected[expected - expected.std() > 0])
self.assertEqual(expected, [it[0] for it in result])
def testReduction(self):
data = self._gen_data(rows=5, value_range=(-100, 100))
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
methods_to_fields = [
(lambda s: df.id.mean(), self.expr.id.mean()),
(lambda s: len(df), self.expr.count()),
(lambda s: df.id.var(ddof=0), self.expr.id.var(ddof=0)),
(lambda s: df.id.std(ddof=0), self.expr.id.std(ddof=0)),
(lambda s: df.id.median(), self.expr.id.median()),
(lambda s: df.id.sum(), self.expr.id.sum()),
(lambda s: df.id.min(), self.expr.id.min()),
(lambda s: df.id.max(), self.expr.id.max()),
(lambda s: df.isMale.min(), self.expr.isMale.min()),
(lambda s: df.name.max(), self.expr.name.max()),
(lambda s: df.birth.max(), self.expr.birth.max()),
(lambda s: df.isMale.sum(), self.expr.isMale.sum()),
(lambda s: df.isMale.any(), self.expr.isMale.any()),
(lambda s: df.isMale.all(), self.expr.isMale.all()),
(lambda s: df.name.nunique(), self.expr.name.nunique()),
]
fields = [it[1].rename('f'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
second = [it[i] for it in result][0]
if isinstance(first, float):
self.assertAlmostEqual(first, second)
else:
self.assertEqual(first, second)
def testMapReduceByApplyDistributeSort(self):
data = [
['name key', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key name', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
def mapper(row):
for word in row[0].split():
yield word, 1
class reducer(object):
def __init__(self):
self._curr = None
self._cnt = 0
def __call__(self, row):
if self._curr is None:
self._curr = row.word
elif self._curr != row.word:
yield (self._curr, self._cnt)
self._curr = row.word
self._cnt = 0
self._cnt += row.count
def close(self):
if self._curr is not None:
yield (self._curr, self._cnt)
expr = self.expr['name', ].apply(
mapper, axis=1, names=['word', 'count'], types=['string', 'int'])
expr = expr.groupby('word').sort('word').apply(
reducer, names=['word', 'count'], types=['string', 'int'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
def testMapReduce(self):
data = [
['name key', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key name', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output(['word', 'cnt'], ['string', 'int'])
def mapper(row):
for word in row[0].split():
yield word, 1
@output(['word', 'cnt'], ['string', 'int'])
def reducer(keys):
cnt = [0, ]
def h(row, done):
cnt[0] += row[1]
if done:
yield keys[0], cnt[0]
return h
expr = self.expr['name', ].map_reduce(mapper, reducer, group='word')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
@output(['word', 'cnt'], ['string', 'int'])
class reducer2(object):
def __init__(self, keys):
self.cnt = 0
def __call__(self, row, done):
self.cnt += row.cnt
if done:
yield row.word, self.cnt
expr = self.expr['name', ].map_reduce(mapper, reducer2, group='word')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
def testDistributeSort(self):
data = [
['name', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output_names('name', 'id')
@output_types('string', 'int')
class reducer(object):
def __init__(self):
self._curr = None
self._cnt = 0
def __call__(self, row):
if self._curr is None:
self._curr = row.name
elif self._curr != row.name:
yield (self._curr, self._cnt)
self._curr = row.name
self._cnt = 0
self._cnt += 1
def close(self):
if self._curr is not None:
yield (self._curr, self._cnt)
expr = self.expr['name', ].groupby('name').sort('name').apply(reducer)
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 2], ['name', 3]]
self.assertEqual(sorted(expected), sorted(result))
def testJoin(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2, _schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [
['name1', 4, -1],
['name2', 1, -2]
]
self.odps.write_table(table2, 0, [table2.new_record(values=d) for d in data2])
try:
expr = self.expr.join(expr2)['name', 'id2']
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 5)
expected = [
[to_str('name1'), 4],
[to_str('name2'), 1]
]
self.assertTrue(all(it in expected for it in result))
expr = self.expr.join(expr2, on=['name', ('id', 'id2')])[self.expr.name, expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 2)
expected = [to_str('name1'), 4]
self.assertTrue(all(it == expected for it in result))
finally:
table2.drop()
def testUnion(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2, _schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [
['name3', 5, -1],
['name4', 6, -2]
]
self.odps.write_table(table2, 0, [table2.new_record(values=d) for d in data2])
try:
expr = self.expr['name', 'id'].distinct().union(expr2[expr2.id2.rename('id'), 'name'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 4],
['name1', 3],
['name2', 2],
['name3', 5],
['name4', 6]
]
result = sorted(result)
expected = sorted(expected)
self.assertEqual(len(result), len(expected))
for e, r in zip(result, expected):
self.assertEqual([to_str(t) for t in e],
[to_str(t) for t in r])
finally:
table2.drop()
def testPersist(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
table_name = 'pyodps_test_engine_persist_table'
try:
df = self.engine.persist(self.expr, table_name)
res = self.engine.execute(df)
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, result)
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
schema = Schema.from_lists(self.schema.names, self.schema.types, ['ds'], ['string'])
self.odps.create_table(table_name, schema)
df = self.engine.persist(self.expr, table_name, partition='ds=today', create_partition=True)
res = self.engine.execute(df)
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, [d[:-1] for d in result])
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
self.engine.persist(self.expr, table_name, partitions=['name'])
t = self.odps.get_table(table_name)
self.assertEqual(2, len(list(t.partitions)))
with t.open_reader(partition='name=name1', reopen=True) as r:
self.assertEqual(4, r.count)
with t.open_reader(partition='name=name2', reopen=True) as r:
self.assertEqual(1, r.count)
finally:
self.odps.delete_table(table_name, if_exists=True)
def teardown(self):
self.table.drop()
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(
['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal',
'datetime'))
self.schema = df_schema_to_odps_schema(schema)
table_name = 'pyodps_test_engine_table'
self.odps.delete_table(table_name, if_exists=True)
self.table = self.odps.create_table(name='pyodps_test_engine_table',
schema=self.schema)
self.expr = CollectionExpr(_source_data=self.table, _schema=schema)
self.engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def _gen_data(self,
rows=None,
data=None,
nullable_field=None,
value_range=None):
if data is None:
data = []
for _ in range(rows):
record = []
for t in self.schema.types:
method = getattr(self, '_gen_random_%s' % t.name)
if t.name == 'bigint':
record.append(method(value_range=value_range))
else:
record.append(method())
data.append(record)
if nullable_field is not None:
j = self.schema._name_indexes[nullable_field]
for i, l in enumerate(data):
if i % 2 == 0:
data[i][j] = None
self.odps.write_table(self.table, 0,
[self.table.new_record(values=d) for d in data])
return data
def testTunnelCases(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr.count()
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(10, result)
expr = self.expr.name.count()
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(10, result)
res = self.engine._handle_cases(self.expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(data, result)
expr = self.expr['name', self.expr.id.rename('new_id')]
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual([it[:2] for it in data], result)
table_name = 'pyodps_test_engine_partitioned'
self.odps.delete_table(table_name, if_exists=True)
df = self.engine.persist(self.expr, table_name, partitions=['name'])
try:
expr = df.count()
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertIsNone(res)
expr = df[df.name == data[0][0]]['fid', 'id'].count()
expr = self.engine._pre_process(expr)
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(res, 0)
expr = df[df.name == data[0][0]]['fid', 'id']
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(len(res), 0)
finally:
self.odps.delete_table(table_name, if_exists=True)
df = self.engine.persist(self.expr,
table_name,
partitions=['name', 'id'])
try:
expr = df.count()
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertIsNone(res)
expr = df[(df.name == data[0][0])
& (df.id == data[0][1])]['fid', 'ismale'].count()
expr = self.engine._pre_process(expr)
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(res, 0)
expr = df[(df.name == data[0][0])
& (df.id == data[0][1])]['fid', 'ismale']
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(len(res), 0)
finally:
self.odps.delete_table(table_name, if_exists=True)
def testAsync(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr.id.sum()
res = self.engine.execute(expr, async=True)
self.assertNotEqual(res.instance.status, Instance.Status.TERMINATED)
res.wait()
self.assertEqual(sum(it[1] for it in data), res.fetch())
def testBase(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr[self.expr.id < 10]['name', lambda x: x.id]
result = self._get_result(self.engine.execute(expr).values)
self.assertEqual(len([it for it in data if it[1] < 10]), len(result))
if len(result) > 0:
self.assertEqual(2, len(result[0]))
expr = self.expr[Scalar(3).rename('const'), self.expr.id,
(self.expr.id + 1).rename('id2')]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual([c.name for c in res.columns], ['const', 'id', 'id2'])
self.assertTrue(all(it[0] == 3 for it in result))
self.assertEqual(len(data), len(result))
self.assertEqual([it[1] + 1 for it in data], [it[2] for it in result])
expr = self.expr.sort('id')[:5]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual(sorted(data, key=lambda it: it[1])[:5], result)
expr = self.expr.sort('id')[:5]
# test do not use tunnel
res = self.engine.execute(expr, use_tunnel=False)
result = self._get_result(res.values)
self.assertEqual(sorted(data, key=lambda it: it[1])[:5], result)
def testElement(self):
data = self._gen_data(5, nullable_field='name')
fields = [
self.expr.name.isnull().rename('name1'),
self.expr.name.notnull().rename('name2'),
self.expr.name.fillna('test').rename('name3'),
self.expr.id.isin([1, 2, 3]).rename('id1'),
self.expr.id.isin(self.expr.fid.astype('int')).rename('id2'),
self.expr.id.notin([1, 2, 3]).rename('id3'),
self.expr.id.notin(self.expr.fid.astype('int')).rename('id4'),
self.expr.id.between(self.expr.fid, 3).rename('id5'),
self.expr.name.fillna('test').switch(
'test',
'test' + self.expr.name.fillna('test'),
'test2',
'test2' + self.expr.name.fillna('test'),
default=self.expr.name).rename('name4'),
self.expr.id.cut([100, 200, 300],
labels=['xsmall', 'small', 'large', 'xlarge'],
include_under=True,
include_over=True).rename('id6')
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual(len([it for it in data if it[0] is None]),
len([it[0] for it in result if it[0]]))
self.assertEqual(len([it[0] for it in data if it[0] is not None]),
len([it[1] for it in result if it[1]]))
self.assertEqual([(it[0] if it[0] is not None else 'test')
for it in data], [it[2] for it in result])
self.assertEqual([(it[1] in (1, 2, 3)) for it in data],
[it[3] for it in result])
fids = [int(it[2]) for it in data]
self.assertEqual([(it[1] in fids) for it in data],
[it[4] for it in result])
self.assertEqual([(it[1] not in (1, 2, 3)) for it in data],
[it[5] for it in result])
self.assertEqual([(it[1] not in fids) for it in data],
[it[6] for it in result])
self.assertEqual([(it[2] <= it[1] <= 3) for it in data],
[it[7] for it in result])
self.assertEqual(
[to_str('testtest' if it[0] is None else it[0]) for it in data],
[to_str(it[8]) for it in result])
def get_val(val):
if val <= 100:
return 'xsmall'
elif 100 < val <= 200:
return 'small'
elif 200 < val <= 300:
return 'large'
else:
return 'xlarge'
self.assertEqual([to_str(get_val(it[1])) for it in data],
[to_str(it[9]) for it in result])
def testArithmetic(self):
data = self._gen_data(5, value_range=(-1000, 1000))
fields = [
(self.expr.id + 1).rename('id1'),
(self.expr.fid - 1).rename('fid1'),
(self.expr.scale * 2).rename('scale1'),
(self.expr.scale + self.expr.id).rename('scale2'),
(self.expr.id / 2).rename('id2'),
(self.expr.id**-2).rename('id3'),
abs(self.expr.id).rename('id4'),
(~self.expr.id).rename('id5'),
(-self.expr.fid).rename('fid2'),
(~self.expr.isMale).rename('isMale1'),
(-self.expr.isMale).rename('isMale2'),
(self.expr.id // 2).rename('id6'),
(self.expr.birth + day(1).rename('birth1')),
(self.expr.birth -
(self.expr.birth - millisecond(10))).rename('birth2'),
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual([it[1] + 1 for it in data], [it[0] for it in result])
self.assertAlmostEqual([it[2] - 1 for it in data],
[it[1] for it in result])
self.assertEqual([it[4] * 2 for it in data], [it[2] for it in result])
self.assertEqual([it[4] + it[1] for it in data],
[it[3] for it in result])
self.assertAlmostEqual([float(it[1]) / 2 for it in data],
[it[4] for it in result])
self.assertEqual([int(it[1]**-2) for it in data],
[it[5] for it in result])
self.assertEqual([abs(it[1]) for it in data], [it[6] for it in result])
self.assertEqual([~it[1] for it in data], [it[7] for it in result])
self.assertAlmostEqual([-it[2] for it in data],
[it[8] for it in result])
self.assertEqual([not it[3] for it in data], [it[9] for it in result])
self.assertEqual([it[1] // 2 for it in data],
[it[11] for it in result])
self.assertEqual([it[5] + timedelta(days=1) for it in data],
[it[12] for it in result])
self.assertEqual([10] * len(data), [it[13] for it in result])
def testMath(self):
data = self._gen_data(5, value_range=(1, 90))
import numpy as np
methods_to_fields = [
(np.sin, self.expr.id.sin()),
(np.cos, self.expr.id.cos()),
(np.tan, self.expr.id.tan()),
(np.sinh, self.expr.id.sinh()),
(np.cosh, self.expr.id.cosh()),
(np.tanh, self.expr.id.tanh()),
(np.log, self.expr.id.log()),
(np.log2, self.expr.id.log2()),
(np.log10, self.expr.id.log10()),
(np.log1p, self.expr.id.log1p()),
(np.exp, self.expr.id.exp()),
(np.expm1, self.expr.id.expm1()),
(np.arccosh, self.expr.id.arccosh()),
(np.arcsinh, self.expr.id.arcsinh()),
(np.arctanh, self.expr.id.arctanh()),
(np.arctan, self.expr.id.arctan()),
(np.sqrt, self.expr.id.sqrt()),
(np.abs, self.expr.id.abs()),
(np.ceil, self.expr.id.ceil()),
(np.floor, self.expr.id.floor()),
(np.trunc, self.expr.id.trunc()),
]
fields = [
it[1].rename('id' + str(i))
for i, it in enumerate(methods_to_fields)
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[1]) for it in data]
second = [it[i] for it in result]
self.assertEqual(len(first), len(second))
for it1, it2 in zip(first, second):
if np.isnan(it1) and np.isnan(it2):
continue
self.assertAlmostEqual(it1, it2)
def testString(self):
data = self._gen_data(5)
methods_to_fields = [
(lambda s: s.capitalize(), self.expr.name.capitalize()),
(lambda s: data[0][0] in s,
self.expr.name.contains(data[0][0], regex=False)),
(lambda s: s.count(data[0][0]), self.expr.name.count(data[0][0])),
(lambda s: s.endswith(data[0][0]),
self.expr.name.endswith(data[0][0])),
(lambda s: s.startswith(data[0][0]),
self.expr.name.startswith(data[0][0])),
(lambda s: s.find(data[0][0]), self.expr.name.find(data[0][0])),
(lambda s: s.rfind(data[0][0]), self.expr.name.rfind(data[0][0])),
(lambda s: s.replace(data[0][0], 'test'),
self.expr.name.replace(data[0][0], 'test')),
(lambda s: s[0], self.expr.name.get(0)),
(lambda s: len(s), self.expr.name.len()),
(lambda s: s.ljust(10), self.expr.name.ljust(10)),
(lambda s: s.ljust(20, '*'), self.expr.name.ljust(20,
fillchar='*')),
(lambda s: s.rjust(10), self.expr.name.rjust(10)),
(lambda s: s.rjust(20, '*'), self.expr.name.rjust(20,
fillchar='*')),
(lambda s: s * 4, self.expr.name.repeat(4)),
(lambda s: s[2:10:2], self.expr.name.slice(2, 10, 2)),
(lambda s: s[-5:-1], self.expr.name.slice(-5, -1)),
(lambda s: s.title(), self.expr.name.title()),
(lambda s: s.rjust(20, '0'), self.expr.name.zfill(20)),
(lambda s: s.isalnum(), self.expr.name.isalnum()),
(lambda s: s.isalpha(), self.expr.name.isalpha()),
(lambda s: s.isdigit(), self.expr.name.isdigit()),
(lambda s: s.isspace(), self.expr.name.isspace()),
(lambda s: s.isupper(), self.expr.name.isupper()),
(lambda s: s.istitle(), self.expr.name.istitle()),
(lambda s: to_str(s).isnumeric(), self.expr.name.isnumeric()),
(lambda s: to_str(s).isdecimal(), self.expr.name.isdecimal()),
]
fields = [
it[1].rename('id' + str(i))
for i, it in enumerate(methods_to_fields)
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[0]) for it in data]
second = [it[i] for it in result]
self.assertEqual(first, second)
def testApply(self):
data = self._gen_data(5)
def my_func(row):
return row.name,
expr = self.expr['name', 'id'].apply(my_func, axis=1, names='name')
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([r[0] for r in result], [r[0] for r in data])
def my_func2(row):
yield len(row.name)
yield row.id
expr = self.expr['name', 'id'].apply(my_func2,
axis=1,
names='cnt',
types='int')
res = self.engine.execute(expr)
result = self._get_result(res)
def gen_expected(data):
for r in data:
yield len(r[0])
yield r[1]
self.assertEqual([r[0] for r in result],
[r for r in gen_expected(data)])
def testDatetime(self):
data = self._gen_data(5)
import pandas as pd
methods_to_fields = [
(lambda s: list(s.birth.dt.year.values), self.expr.birth.year),
(lambda s: list(s.birth.dt.month.values), self.expr.birth.month),
(lambda s: list(s.birth.dt.day.values), self.expr.birth.day),
(lambda s: list(s.birth.dt.hour.values), self.expr.birth.hour),
(lambda s: list(s.birth.dt.minute.values), self.expr.birth.minute),
(lambda s: list(s.birth.dt.second.values), self.expr.birth.second),
(lambda s: list(s.birth.dt.weekofyear.values),
self.expr.birth.weekofyear),
(lambda s: list(s.birth.dt.dayofweek.values),
self.expr.birth.dayofweek),
(lambda s: list(s.birth.dt.weekday.values),
self.expr.birth.weekday),
(lambda s: list(s.birth.dt.date.values), self.expr.birth.date),
(lambda s: list(s.birth.dt.strftime('%Y%d')),
self.expr.birth.strftime('%Y%d'))
]
fields = [
it[1].rename('birth' + str(i))
for i, it in enumerate(methods_to_fields)
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
def conv(v):
if isinstance(v, pd.Timestamp):
return v.to_datetime().date()
else:
return v
second = [conv(it[i]) for it in result]
self.assertEqual(first, second)
def testSortDistinct(self):
data = [
['name1', 4, None, None, None, None],
['name2', 2, None, None, None, None],
['name1', 4, None, None, None, None],
['name1', 3, None, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.sort(['name', -self.expr.id
]).distinct(['name', lambda x: x.id + 1])[:50]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 3)
expected = [['name1', 5], ['name1', 4], ['name2', 3]]
self.assertEqual(expected, result)
def testGroupbyAggregation(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(['name', 'id'])[lambda x: x.fid.min() * 2 < 8] \
.agg(self.expr.fid.max() + 1, new_id=self.expr.id.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['name1', 3, 5.1, 6], ['name2', 2, 4.5, 2]]
result = sorted(result, key=lambda k: k[0])
self.assertEqual(expected, result)
field = self.expr.groupby('name').sort(['id',
-self.expr.fid]).row_number()
expr = self.expr['name', 'id', 'fid', field]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 4.1, 1],
['name1', 3, 2.2, 2],
['name1', 4, 5.3, 3],
['name1', 4, 4.2, 4],
['name2', 2, 3.5, 1],
]
result = sorted(result, key=lambda k: (k[0], k[1], -k[2]))
self.assertEqual(expected, result)
expr = self.expr.name.value_counts()[:25]
expected = [['name1', 4], ['name2', 1]]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.name.topk(25)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.groupby('name').count()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([it[1:] for it in expected], result)
expected = [['name1', 2], ['name2', 1]]
expr = self.expr.groupby('name').id.nunique()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([it[1:] for it in expected], result)
expr = self.expr[self.expr['id'] > 2].name.value_counts()[:25]
expected = [['name1', 4]]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
def testJoinGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2,
_schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [['name1', 4, -1], ['name2', 1, -2]]
self.odps.write_table(table2, 0,
[table2.new_record(values=d) for d in data2])
expr = self.expr.join(expr2, on='name')[self.expr]
expr = expr.groupby('id').agg(expr.fid.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
import pandas as pd
expected = pd.DataFrame(data, columns=self.expr.schema.names).groupby('id').agg({'fid': 'sum'})\
.reset_index().values.tolist()
for it in zip(sorted(expected, key=lambda it: it[0]),
sorted(result, key=lambda it: it[0])):
self.assertAlmostEqual(it[0][0], it[1][0])
self.assertAlmostEqual(it[0][1], it[1][1])
def testFilterGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(
['name']).agg(id=self.expr.id.max())[lambda x: x.id > 3]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 1)
expected = [['name1', 4]]
self.assertEqual(expected, result)
def testWindowRewrite(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr[self.expr.id - self.expr.id.mean() < 10][[
lambda x: x.id - x.id.max()
]][[lambda x: x.id - x.id.min()]][lambda x: x.id - x.id.std() > 0]
res = self.engine.execute(expr)
result = self._get_result(res)
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
expected = df.id - df.id.max()
expected = expected - expected.min()
expected = list(expected[expected - expected.std() > 0])
self.assertEqual(expected, [it[0] for it in result])
def testReduction(self):
data = self._gen_data(rows=5, value_range=(-100, 100))
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
methods_to_fields = [
(lambda s: df.id.mean(), self.expr.id.mean()),
(lambda s: len(df), self.expr.count()),
(lambda s: df.id.var(ddof=0), self.expr.id.var(ddof=0)),
(lambda s: df.id.std(ddof=0), self.expr.id.std(ddof=0)),
(lambda s: df.id.median(), self.expr.id.median()),
(lambda s: df.id.sum(), self.expr.id.sum()),
(lambda s: df.id.min(), self.expr.id.min()),
(lambda s: df.id.max(), self.expr.id.max()),
(lambda s: df.isMale.min(), self.expr.isMale.min()),
(lambda s: df.name.max(), self.expr.name.max()),
(lambda s: df.birth.max(), self.expr.birth.max()),
(lambda s: df.isMale.sum(), self.expr.isMale.sum()),
(lambda s: df.isMale.any(), self.expr.isMale.any()),
(lambda s: df.isMale.all(), self.expr.isMale.all()),
(lambda s: df.name.nunique(), self.expr.name.nunique()),
]
fields = [
it[1].rename('f' + str(i))
for i, it in enumerate(methods_to_fields)
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
second = [it[i] for it in result][0]
if isinstance(first, float):
self.assertAlmostEqual(first, second)
else:
self.assertEqual(first, second)
def testMapReduceByApplyDistributeSort(self):
data = [
['name key', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key name', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
def mapper(row):
for word in row[0].split():
yield word, 1
class reducer(object):
def __init__(self):
self._curr = None
self._cnt = 0
def __call__(self, row):
if self._curr is None:
self._curr = row.word
elif self._curr != row.word:
yield (self._curr, self._cnt)
self._curr = row.word
self._cnt = 0
self._cnt += row.count
def close(self):
if self._curr is not None:
yield (self._curr, self._cnt)
expr = self.expr['name', ].apply(mapper,
axis=1,
names=['word', 'count'],
types=['string', 'int'])
expr = expr.groupby('word').sort('word').apply(reducer,
names=['word', 'count'],
types=['string', 'int'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
def testMapReduce(self):
data = [
['name key', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key name', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output(['word', 'cnt'], ['string', 'int'])
def mapper(row):
for word in row[0].split():
yield word, 1
@output(['word', 'cnt'], ['string', 'int'])
def reducer(keys):
cnt = [
0,
]
def h(row, done):
cnt[0] += row[1]
if done:
yield keys[0], cnt[0]
return h
expr = self.expr['name', ].map_reduce(mapper, reducer, group='word')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
@output(['word', 'cnt'], ['string', 'int'])
class reducer2(object):
def __init__(self, keys):
self.cnt = 0
def __call__(self, row, done):
self.cnt += row.cnt
if done:
yield row.word, self.cnt
expr = self.expr['name', ].map_reduce(mapper, reducer2, group='word')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
def testDistributeSort(self):
data = [
['name', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output_names('name', 'id')
@output_types('string', 'int')
class reducer(object):
def __init__(self):
self._curr = None
self._cnt = 0
def __call__(self, row):
if self._curr is None:
self._curr = row.name
elif self._curr != row.name:
yield (self._curr, self._cnt)
self._curr = row.name
self._cnt = 0
self._cnt += 1
def close(self):
if self._curr is not None:
yield (self._curr, self._cnt)
expr = self.expr['name', ].groupby('name').sort('name').apply(reducer)
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 2], ['name', 3]]
self.assertEqual(sorted(expected), sorted(result))
def testJoin(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2,
_schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [['name1', 4, -1], ['name2', 1, -2]]
self.odps.write_table(table2, 0,
[table2.new_record(values=d) for d in data2])
try:
expr = self.expr.join(expr2)['name', 'id2']
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 5)
expected = [[to_str('name1'), 4], [to_str('name2'), 1]]
self.assertTrue(all(it in expected for it in result))
expr = self.expr.join(expr2, on=['name',
('id', 'id2')])[self.expr.name,
expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 2)
expected = [to_str('name1'), 4]
self.assertTrue(all(it == expected for it in result))
finally:
table2.drop()
def testUnion(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2,
_schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [['name3', 5, -1], ['name4', 6, -2]]
self.odps.write_table(table2, 0,
[table2.new_record(values=d) for d in data2])
try:
expr = self.expr['name', 'id'].distinct().union(
expr2[expr2.id2.rename('id'), 'name'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['name1', 4], ['name1', 3], ['name2', 2], ['name3', 5],
['name4', 6]]
result = sorted(result)
expected = sorted(expected)
self.assertEqual(len(result), len(expected))
for e, r in zip(result, expected):
self.assertEqual([to_str(t) for t in e],
[to_str(t) for t in r])
finally:
table2.drop()
def testPersist(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
table_name = 'pyodps_test_engine_persist_table'
try:
df = self.engine.persist(self.expr, table_name)
res = self.engine.execute(df)
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, result)
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
schema = Schema.from_lists(self.schema.names, self.schema.types,
['ds'], ['string'])
self.odps.create_table(table_name, schema)
df = self.engine.persist(self.expr,
table_name,
partition='ds=today',
create_partition=True)
res = self.engine.execute(df)
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, [d[:-1] for d in result])
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
self.engine.persist(self.expr, table_name, partitions=['name'])
t = self.odps.get_table(table_name)
self.assertEqual(2, len(list(t.partitions)))
with t.open_reader(partition='name=name1', reopen=True) as r:
self.assertEqual(4, r.count)
with t.open_reader(partition='name=name2', reopen=True) as r:
self.assertEqual(1, r.count)
finally:
self.odps.delete_table(table_name, if_exists=True)
def teardown(self):
self.table.drop()
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid'],
datatypes('string', 'int64', 'float64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
self.engine = ODPSEngine(self.odps)
def testSimpleLambda(self):
self.engine.compile(self.expr.id.map(lambda x: x + 1))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([4, ], runners.simple_run(udf, [(3, ), ]))
def testSimpleFunction(self):
def my_func(x):
if x < 0:
return -1
elif x == 0:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1], runners.simple_run(udf, [(-3, ), (0, ), (5, )]))
def testNestFunction(self):
def my_func(x):
def inner(y):
if y < 0:
return -2
elif y == 0:
return 0
else:
return 2
return inner(x)
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-2, 0, 2], runners.simple_run(udf, [(-3, ), (0, ), (5, )]))
def testGlobalVarFunction(self):
global_val = 10
def my_func(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1], runners.simple_run(udf, [(-9, ), (10, ), (15, )]))
def testRefFuncFunction(self):
global_val = 10
def my_func1(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
def my_func(y):
return my_func1(y)
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1], runners.simple_run(udf, [(-9, ), (10, ), (15, )]))
def testApplyToSequenceFuntion(self):
def my_func(row):
return row.name + str(row.id)
self.engine.compile(self.expr.apply(my_func, axis=1, reduce=True).rename('test'))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertEqual(['name1', 'name2'],
runners.simple_run(udf, [('name', 1, None), ('name', 2, None)]))
def testApplyFunction(self):
def my_func(row):
return row.name, row.id
self.engine.compile(self.expr.apply(my_func, axis=1, names=['name', 'id'], types=['string', 'int']))
udtf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udtf, globals(), locals())
udtf = locals()[UDF_CLASS_NAME]
self.assertEqual([('name1', 1), ('name2', 2)],
runners.simple_run(udtf, [('name1', 1, None), ('name2', 2, None)]))
def testApplyGeneratorFunction(self):
def my_func(row):
for n in row.name.split(','):
yield n
self.engine.compile(self.expr.apply(my_func, axis=1, names='name'))
udtf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udtf, globals(), locals())
udtf = locals()[UDF_CLASS_NAME]
self.assertEqual(['name1', 'name2', 'name3', 'name4'],
runners.simple_run(udtf, [('name1,name2', 1, None), ('name3,name4', 2, None)]))
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid'],
datatypes('string', 'int64', 'float64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
self.engine = ODPSEngine(self.odps)
def testSimpleLambda(self):
self.engine.compile(self.expr.id.map(lambda x: x + 1))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([4, ], runners.simple_run(udf, [(3, ), ]))
def testSimpleFunction(self):
def my_func(x):
if x < 0:
return -1
elif x == 0:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1], runners.simple_run(udf, [(-3, ), (0, ), (5, )]))
def testNestFunction(self):
def my_func(x):
def inner(y):
if y < 0:
return -2
elif y == 0:
return 0
else:
return 2
return inner(x)
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-2, 0, 2], runners.simple_run(udf, [(-3, ), (0, ), (5, )]))
def testGlobalVarFunction(self):
global_val = 10
def my_func(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1], runners.simple_run(udf, [(-9, ), (10, ), (15, )]))
def testRefFuncFunction(self):
global_val = 10
def my_func1(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
def my_func(y):
return my_func1(y)
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1], runners.simple_run(udf, [(-9, ), (10, ), (15, )]))
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal', 'datetime'))
self.schema = df_schema_to_odps_schema(schema)
table_name = 'pyodps_test_engine_table'
self.odps.delete_table(table_name, if_exists=True)
self.table = self.odps.create_table(
name='pyodps_test_engine_table', schema=self.schema)
self.expr = CollectionExpr(_source_data=self.table, _schema=schema)
self.engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def _gen_random_bigint(self, value_range=None):
return random.randint(*(value_range or types.bigint._bounds))
def _gen_random_string(self, max_length=15):
gen_letter = lambda: letters[random.randint(0, 51)]
return to_str(''.join([gen_letter() for _ in range(random.randint(1, 15))]))
def _gen_random_double(self):
return random.uniform(-2**32, 2**32)
def _gen_random_datetime(self):
return datetime.fromtimestamp(random.randint(0, int(time.time())))
def _gen_random_boolean(self):
return random.uniform(-1, 1) > 0
def _gen_random_decimal(self):
return Decimal(str(self._gen_random_double()))
def _gen_data(self, rows=None, data=None, nullable_field=None, value_range=None):
if data is None:
data = []
for _ in range(rows):
record = []
for t in self.schema.types:
method = getattr(self, '_gen_random_%s' % t.name)
if t.name == 'bigint':
record.append(method(value_range=value_range))
else:
record.append(method())
data.append(record)
if nullable_field is not None:
j = self.schema._name_indexes[nullable_field]
for i, l in enumerate(data):
if i % 2 == 0:
data[i][j] = None
self.odps.write_table(self.table, 0, [self.table.new_record(values=d) for d in data])
return data
def _get_result(self, res):
if isinstance(res, ResultFrame):
res = res.values
try:
import pandas
if isinstance(res, pandas.DataFrame):
return [list(it) for it in res.values]
else:
return res
except ImportError:
return res
def testTunnelCases(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr.count()
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(10, result)
expr = self.expr.name.count()
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(10, result)
res = self.engine._handle_cases(self.expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual(data, result)
expr = self.expr['name', self.expr.id.rename('new_id')]
res = self.engine._handle_cases(expr, self.faked_bar)
result = self._get_result(res)
self.assertEqual([it[:2] for it in data], result)
table_name = 'pyodps_test_engine_partitioned'
self.odps.delete_table(table_name, if_exists=True)
df = self.expr.persist(table_name, partitions=['name'])
try:
expr = df.count()
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertIsNone(res)
expr = df[df.name == data[0][0]]['fid', 'id'].count()
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(res, 0)
expr = df[df.name == data[0][0]]['fid', 'id']
res = self.engine._handle_cases(expr, self.faked_bar)
self.assertGreater(len(res), 0)
finally:
self.odps.delete_table(table_name, if_exists=True)
def testBase(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr[self.expr.id < 10]['name', lambda x: x.id]
result = self._get_result(self.engine.execute(expr).values)
self.assertEqual(len([it for it in data if it[1] < 10]), len(result))
if len(result) > 0:
self.assertEqual(2, len(result[0]))
expr = self.expr[Scalar(3).rename('const'), self.expr.id, (self.expr.id + 1).rename('id2')]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual([c.name for c in res.columns], ['const', 'id', 'id2'])
self.assertTrue(all(it[0] == 3 for it in result))
self.assertEqual(len(data), len(result))
self.assertEqual([it[1]+1 for it in data], [it[2] for it in result])
expr = self.expr.sort('id')[:5]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual(sorted(data, key=lambda it: it[1])[:5], result)
def testElement(self):
data = self._gen_data(5, nullable_field='name')
fields = [
self.expr.name.isnull().rename('name1'),
self.expr.name.notnull().rename('name2'),
self.expr.name.fillna('test').rename('name3'),
self.expr.id.isin([1, 2, 3]).rename('id1'),
self.expr.id.isin(self.expr.fid.astype('int')).rename('id2'),
self.expr.id.notin([1, 2, 3]).rename('id3'),
self.expr.id.notin(self.expr.fid.astype('int')).rename('id4'),
self.expr.id.between(self.expr.fid, 3).rename('id5'),
self.expr.name.fillna('test').switch('test', 'test' + self.expr.name.fillna('test'),
'test2', 'test2' + self.expr.name.fillna('test'),
default=self.expr.name).rename('name4'),
self.expr.id.cut([100, 200, 300],
labels=['xsmall', 'small', 'large', 'xlarge'],
include_under=True, include_over=True).rename('id6')
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual(len([it for it in data if it[0] is None]),
len([it[0] for it in result if it[0]]))
self.assertEqual(len([it[0] for it in data if it[0] is not None]),
len([it[1] for it in result if it[1]]))
self.assertEqual([(it[0] if it[0] is not None else 'test') for it in data],
[it[2] for it in result])
self.assertEqual([(it[1] in (1, 2, 3)) for it in data],
[it[3] for it in result])
fids = [int(it[2]) for it in data]
self.assertEqual([(it[1] in fids) for it in data],
[it[4] for it in result])
self.assertEqual([(it[1] not in (1, 2, 3)) for it in data],
[it[5] for it in result])
self.assertEqual([(it[1] not in fids) for it in data],
[it[6] for it in result])
self.assertEqual([(it[2] <= it[1] <= 3) for it in data],
[it[7] for it in result])
self.assertEqual([to_str('testtest' if it[0] is None else it[0]) for it in data],
[to_str(it[8]) for it in result])
def get_val(val):
if val <= 100:
return 'xsmall'
elif 100 < val <= 200:
return 'small'
elif 200 < val <= 300:
return 'large'
else:
return 'xlarge'
self.assertEqual([to_str(get_val(it[1])) for it in data], [to_str(it[9]) for it in result])
def testArithmetic(self):
data = self._gen_data(5, value_range=(-1000, 1000))
fields = [
(self.expr.id + 1).rename('id1'),
(self.expr.fid - 1).rename('fid1'),
(self.expr.scale * 2).rename('scale1'),
(self.expr.scale + self.expr.id).rename('scale2'),
(self.expr.id / 2).rename('id2'),
(self.expr.id ** -2).rename('id3'),
abs(self.expr.id).rename('id4'),
(~self.expr.id).rename('id5'),
(-self.expr.fid).rename('fid2'),
(~self.expr.isMale).rename('isMale1'),
(-self.expr.isMale).rename('isMale2'),
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual([it[1] + 1 for it in data],
[it[0] for it in result])
self.assertAlmostEqual([it[2] - 1 for it in data],
[it[1] for it in result])
self.assertEqual([it[4] * 2 for it in data],
[it[2] for it in result])
self.assertEqual([it[4] + it[1] for it in data],
[it[3] for it in result])
self.assertAlmostEqual([float(it[1]) / 2 for it in data],
[it[4] for it in result])
self.assertEqual([int(it[1] ** -2) for it in data],
[it[5] for it in result])
self.assertEqual([abs(it[1]) for it in data],
[it[6] for it in result])
self.assertEqual([~it[1] for it in data],
[it[7] for it in result])
self.assertAlmostEqual([-it[2] for it in data],
[it[8] for it in result])
self.assertEqual([not it[3] for it in data],
[it[9] for it in result])
# TODO: test the datetime add and substract
def testMath(self):
data = self._gen_data(5, value_range=(1, 90))
import numpy as np
methods_to_fields = [
(np.sin, self.expr.id.sin()),
(np.cos, self.expr.id.cos()),
(np.tan, self.expr.id.tan()),
(np.sinh, self.expr.id.sinh()),
(np.cosh, self.expr.id.cosh()),
(np.tanh, self.expr.id.tanh()),
(np.log, self.expr.id.log()),
(np.log2, self.expr.id.log2()),
(np.log10, self.expr.id.log10()),
(np.log1p, self.expr.id.log1p()),
(np.exp, self.expr.id.exp()),
(np.expm1, self.expr.id.expm1()),
(np.arccosh, self.expr.id.arccosh()),
(np.arcsinh, self.expr.id.arcsinh()),
(np.arctanh, self.expr.id.arctanh()),
(np.arctan, self.expr.id.arctan()),
(np.sqrt, self.expr.id.sqrt()),
(np.abs, self.expr.id.abs()),
(np.ceil, self.expr.id.ceil()),
(np.floor, self.expr.id.floor()),
(np.trunc, self.expr.id.trunc()),
]
fields = [it[1].rename('id'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[1]) for it in data]
second = [it[i] for it in result]
self.assertEqual(len(first), len(second))
for it1, it2 in zip(first, second):
if np.isnan(it1) and np.isnan(it2):
continue
self.assertAlmostEqual(it1, it2)
def testString(self):
data = self._gen_data(5)
methods_to_fields = [
(lambda s: s.capitalize(), self.expr.name.capitalize()),
(lambda s: data[0][0] in s, self.expr.name.contains(data[0][0], regex=False)),
(lambda s: s.count(data[0][0]), self.expr.name.count(data[0][0])),
(lambda s: s.endswith(data[0][0]), self.expr.name.endswith(data[0][0])),
(lambda s: s.startswith(data[0][0]), self.expr.name.startswith(data[0][0])),
(lambda s: s.find(data[0][0]), self.expr.name.find(data[0][0])),
(lambda s: s.rfind(data[0][0]), self.expr.name.rfind(data[0][0])),
(lambda s: s.replace(data[0][0], 'test'), self.expr.name.replace(data[0][0], 'test')),
(lambda s: s[0], self.expr.name.get(0)),
(lambda s: len(s), self.expr.name.len()),
(lambda s: s.ljust(10), self.expr.name.ljust(10)),
(lambda s: s.ljust(20, '*'), self.expr.name.ljust(20, fillchar='*')),
(lambda s: s.rjust(10), self.expr.name.rjust(10)),
(lambda s: s.rjust(20, '*'), self.expr.name.rjust(20, fillchar='*')),
(lambda s: s * 4, self.expr.name.repeat(4)),
(lambda s: s[2: 10: 2], self.expr.name.slice(2, 10, 2)),
(lambda s: s[-5: -1], self.expr.name.slice(-5, -1)),
(lambda s: s.title(), self.expr.name.title()),
(lambda s: s.rjust(20, '0'), self.expr.name.zfill(20)),
(lambda s: s.isalnum(), self.expr.name.isalnum()),
(lambda s: s.isalpha(), self.expr.name.isalpha()),
(lambda s: s.isdigit(), self.expr.name.isdigit()),
(lambda s: s.isspace(), self.expr.name.isspace()),
(lambda s: s.isupper(), self.expr.name.isupper()),
(lambda s: s.istitle(), self.expr.name.istitle()),
(lambda s: to_str(s).isnumeric(), self.expr.name.isnumeric()),
(lambda s: to_str(s).isdecimal(), self.expr.name.isdecimal()),
]
fields = [it[1].rename('id'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[0]) for it in data]
second = [it[i] for it in result]
self.assertEqual(first, second)
def testDatetime(self):
data = self._gen_data(5)
import pandas as pd
methods_to_fields = [
(lambda s: list(s.birth.dt.year.values), self.expr.birth.year),
(lambda s: list(s.birth.dt.month.values), self.expr.birth.month),
(lambda s: list(s.birth.dt.day.values), self.expr.birth.day),
(lambda s: list(s.birth.dt.hour.values), self.expr.birth.hour),
(lambda s: list(s.birth.dt.minute.values), self.expr.birth.minute),
(lambda s: list(s.birth.dt.second.values), self.expr.birth.second),
(lambda s: list(s.birth.dt.weekofyear.values), self.expr.birth.weekofyear),
(lambda s: list(s.birth.dt.dayofweek.values), self.expr.birth.dayofweek),
(lambda s: list(s.birth.dt.weekday.values), self.expr.birth.weekday),
]
fields = [it[1].rename('birth'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
second = [it[i] for it in result]
self.assertEqual(first, second)
def testSortDistinct(self):
data = [
['name1', 4, None, None, None, None],
['name2', 2, None, None, None, None],
['name1', 4, None, None, None, None],
['name1', 3, None, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.sort(['name', -self.expr.id]).distinct(['name', lambda x: x.id + 1])[:50]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 3)
expected = [
['name1', 5],
['name1', 4],
['name2', 3]
]
self.assertEqual(expected, result)
def testGroupbyAggregation(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(['name', 'id'])[lambda x: x.fid.min() * 2 < 8] \
.agg(self.expr.fid.max() + 1, new_id=self.expr.id.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 5.1, 6],
['name2', 2, 4.5, 2]
]
result = sorted(result, key=lambda k: k[0])
self.assertEqual(expected, result)
field = self.expr.groupby('name').sort(['id', -self.expr.fid]).row_number()
expr = self.expr['name', 'id', 'fid', field]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 4.1, 1],
['name1', 3, 2.2, 2],
['name1', 4, 5.3, 3],
['name1', 4, 4.2, 4],
['name2', 2, 3.5, 1],
]
result = sorted(result, key=lambda k: (k[0], k[1], -k[2]))
self.assertEqual(expected, result)
expr = self.expr.name.value_counts()[:25]
expected = [
['name1', 4],
['name2', 1]
]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.name.topk(25)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.groupby('name').count()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
def testFilterGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(['name']).agg(id=self.expr.id.max())[lambda x: x.id > 3]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 1)
expected = [
['name1', 4]
]
self.assertEqual(expected, result)
def testWindowRewrite(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr[self.expr.id - self.expr.id.mean() < 10][
[lambda x: x.id - x.id.max()]][[lambda x: x.id - x.id.min()]][lambda x: x.id - x.id.std() > 0]
# FIXME compiling too slow
res = self.engine.execute(expr)
result = self._get_result(res)
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
expected = df.id - df.id.max()
expected = expected - expected.min()
expected = list(expected[expected - expected.std() > 0])
self.assertEqual(expected, [it[0] for it in result])
def testReduction(self):
data = self._gen_data(rows=5, value_range=(-100, 100))
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
methods_to_fields = [
(lambda s: df.id.mean(), self.expr.id.mean()),
(lambda s: len(df), self.expr.count()),
(lambda s: df.id.var(ddof=0), self.expr.id.var(ddof=0)),
(lambda s: df.id.std(ddof=0), self.expr.id.std(ddof=0)),
(lambda s: df.id.median(), self.expr.id.median()),
(lambda s: df.id.sum(), self.expr.id.sum()),
(lambda s: df.id.min(), self.expr.id.min()),
(lambda s: df.id.max(), self.expr.id.max()),
(lambda s: df.isMale.min(), self.expr.isMale.min()),
(lambda s: df.name.max(), self.expr.name.max()),
(lambda s: df.birth.max(), self.expr.birth.max()),
(lambda s: df.name.sum(), self.expr.name.sum()),
(lambda s: df.isMale.sum(), self.expr.isMale.sum()),
]
fields = [it[1].rename('f'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
second = [it[i] for it in result][0]
self.assertAlmostEqual(first, second)
def testJoin(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2, _schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [
['name1', 4, -1],
['name2', 1, -2]
]
self.odps.write_table(table2, 0, [table2.new_record(values=d) for d in data2])
try:
expr = self.expr.join(expr2)['name', 'id2']
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 5)
expected = [
[to_str('name1'), 4],
[to_str('name2'), 1]
]
self.assertTrue(all(it in expected for it in result))
expr = self.expr.join(expr2, on=['name', ('id', 'id2')])[self.expr.name, expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 2)
expected = [to_str('name1'), 4]
self.assertTrue(all(it == expected for it in result))
finally:
table2.drop()
def testUnion(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = 'pyodps_test_engine_table2'
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=schema2)
expr2 = CollectionExpr(_source_data=table2, _schema=odps_schema_to_df_schema(schema2))
self._gen_data(data=data)
data2 = [
['name3', 5, -1],
['name4', 6, -2]
]
self.odps.write_table(table2, 0, [table2.new_record(values=d) for d in data2])
try:
expr = self.expr['name', 'id'].distinct().union(expr2[expr2.id2.rename('id'), 'name'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 4],
['name1', 3],
['name2', 2],
['name3', 5],
['name4', 6]
]
result = sorted(result)
expected = sorted(expected)
self.assertEqual(len(result), len(expected))
for e, r in zip(result, expected):
self.assertEqual([to_str(t) for t in e],
[to_str(t) for t in r])
finally:
table2.drop()
def testPersist(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
table_name = 'pyodps_test_engine_persist_table'
try:
df = self.expr.persist(table_name)
res = self.engine.execute(df)
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, result)
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
self.expr.persist(table_name, partitions=['name'])
t = self.odps.get_table(table_name)
self.assertEqual(2, len(list(t.partitions)))
with t.open_reader(partition='name=name1', reopen=True) as r:
self.assertEqual(4, r.count)
with t.open_reader(partition='name=name2', reopen=True) as r:
self.assertEqual(1, r.count)
finally:
self.odps.delete_table(table_name, if_exists=True)
def teardown(self):
self.table.drop()
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid'],
datatypes('string', 'int64', 'float64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
self.engine = ODPSEngine(self.odps)
def testSimpleLambda(self):
self.engine.compile(self.expr.id.map(lambda x: x + 1))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([
4,
], runners.simple_run(udf, [
(3, ),
]))
def testSimpleFunction(self):
def my_func(x):
if x < 0:
return -1
elif x == 0:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1],
runners.simple_run(udf, [(-3, ), (0, ),
(5, )]))
def testNestFunction(self):
def my_func(x):
def inner(y):
if y < 0:
return -2
elif y == 0:
return 0
else:
return 2
return inner(x)
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-2, 0, 2],
runners.simple_run(udf, [(-3, ), (0, ),
(5, )]))
def testGlobalVarFunction(self):
global_val = 10
def my_func(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1],
runners.simple_run(udf, [(-9, ), (10, ),
(15, )]))
def testRefFuncFunction(self):
global_val = 10
def my_func1(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
def my_func(y):
return my_func1(y)
self.engine.compile(self.expr.id.map(my_func))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1],
runners.simple_run(udf, [(-9, ), (10, ),
(15, )]))
def testApplyToSequenceFuntion(self):
def my_func(row):
return row.name + str(row.id)
self.engine.compile(
self.expr.apply(my_func, axis=1, reduce=True).rename('test'))
udf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertEqual(['name1', 'name2'],
runners.simple_run(udf, [('name', 1, None),
('name', 2, None)]))
def testApplyFunction(self):
def my_func(row):
return row.name, row.id
self.engine.compile(
self.expr.apply(my_func,
axis=1,
names=['name', 'id'],
types=['string', 'int']))
udtf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udtf, globals(), locals())
udtf = locals()[UDF_CLASS_NAME]
self.assertEqual([('name1', 1), ('name2', 2)],
runners.simple_run(udtf, [('name1', 1, None),
('name2', 2, None)]))
def testApplyGeneratorFunction(self):
def my_func(row):
for n in row.name.split(','):
yield n
self.engine.compile(self.expr.apply(my_func, axis=1, names='name'))
udtf = list(self.engine._ctx._func_to_udfs.values())[0]
six.exec_(udtf, globals(), locals())
udtf = locals()[UDF_CLASS_NAME]
self.assertEqual(['name1', 'name2', 'name3', 'name4'],
runners.simple_run(udtf, [('name1,name2', 1, None),
('name3,name4', 2, None)]))
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid'],
datatypes('string', 'int64', 'float64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
self.engine = ODPSEngine(self.odps)
def testSimpleLambda(self):
self.engine.compile(self.expr.id.map(lambda x: x + 1))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([
4,
], runners.simple_run(udf, [
(3, ),
]))
def testSimpleFunction(self):
def my_func(x):
if x < 0:
return -1
elif x == 0:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf)
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1],
runners.simple_run(udf, [(-3, ), (0, ),
(5, )]))
def testNestFunction(self):
def my_func(x):
def inner(y):
if y < 0:
return -2
elif y == 0:
return 0
else:
return 2
return inner(x)
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-2, 0, 2],
runners.simple_run(udf, [(-3, ), (0, ),
(5, )]))
def testGlobalVarFunction(self):
global_val = 10
def my_func(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1],
runners.simple_run(udf, [(-9, ), (10, ),
(15, )]))
def testRefFuncFunction(self):
global_val = 10
def my_func1(x):
if x < global_val:
return -1
elif x == global_val:
return 0
else:
return 1
def my_func(y):
return my_func1(y)
self.engine.compile(self.expr.id.map(my_func))
udf = self.engine._ctx._func_to_udfs.values()[0]
exec(udf, globals(), locals())
udf = locals()[UDF_CLASS_NAME]
self.assertSequenceEqual([-1, 0, 1],
runners.simple_run(udf, [(-9, ), (10, ),
(15, )]))
def testFilterPushDownThroughJoin(self):
expr = self.expr.join(self.expr3, on='name')
expr = expr[(expr.id_x < 10) & (expr.fid_y > 3)]
expected = 'SELECT t2.`name`, t2.`id` AS `id_x`, t2.`fid` AS `fid_x`, ' \
't2.`isMale`, t2.`scale`, t2.`birth`, t2.`ds`, t4.`id` AS `id_y`, ' \
't4.`fid` AS `fid_y`, t4.`part1`, t4.`part2` \n' \
'FROM (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
' WHERE t1.`id` < 10\n' \
') t2 \n' \
'INNER JOIN \n' \
' (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table2` t3 \n' \
' WHERE t3.`fid` > 3\n' \
' ) t4\n' \
'ON t2.`name` == t4.`name`'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.join(self.expr3, on='name')
expr = expr[(expr.id_x < 10) & (expr.fid_y > 3) & (expr.id_x > 3)]
expected = 'SELECT t2.`name`, t2.`id` AS `id_x`, t2.`fid` AS `fid_x`, ' \
't2.`isMale`, t2.`scale`, t2.`birth`, t2.`ds`, t4.`id` AS `id_y`, ' \
't4.`fid` AS `fid_y`, t4.`part1`, t4.`part2` \n' \
'FROM (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
' WHERE (t1.`id` < 10) AND (t1.`id` > 3)\n' \
') t2 \n' \
'INNER JOIN \n' \
' (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table2` t3 \n' \
' WHERE t3.`fid` > 3\n' \
' ) t4\n' \
'ON t2.`name` == t4.`name`'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr[self.expr.name, self.expr.id + 1]
expr2 = self.expr3['tt' + self.expr3.name, self.expr3.id.rename('id2')]
expr = expr.join(expr2, on='name')
expr = expr[((expr.id < 10) | (expr.id > 100)) & (expr.id2 > 3)]
expected = "SELECT t2.`name`, t2.`id`, t4.`id2` \n" \
"FROM (\n" \
" SELECT t1.`name`, t1.`id` + 1 AS `id` \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
" WHERE ((t1.`id` + 1) < 10) OR ((t1.`id` + 1) > 100)\n" \
") t2 \n" \
"INNER JOIN \n" \
" (\n" \
" SELECT CONCAT('tt', t3.`name`) AS `name`, t3.`id` AS `id2` \n" \
" FROM mocked_project.`pyodps_test_expr_table2` t3 \n" \
" WHERE t3.`id` > 3\n" \
" ) t4\n" \
"ON t2.`name` == t4.`name`"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.join(self.expr3, on='name')
expr = expr[(expr.id_x + expr.id_y < 10) & (expr.id_x > 3)]
expected = "SELECT * \n" \
"FROM (\n" \
" SELECT t2.`name`, t2.`id` AS `id_x`, t2.`fid` AS `fid_x`, t2.`isMale`, " \
"t2.`scale`, t2.`birth`, t2.`ds`, t3.`id` AS `id_y`, " \
"t3.`fid` AS `fid_y`, t3.`part1`, t3.`part2` \n" \
" FROM (\n" \
" SELECT * \n" \
" FROM mocked_project.`pyodps_test_expr_table` t1 \n" \
" WHERE t1.`id` > 3\n" \
" ) t2 \n" \
" INNER JOIN \n" \
" mocked_project.`pyodps_test_expr_table2` t3\n" \
" ON t2.`name` == t3.`name` \n" \
") t4 \n" \
"WHERE (t4.`id_x` + t4.`id_y`) < 10"
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
expr = self.expr.join(self.expr3,
on=[
'name', self.expr.id == self.expr3.id,
self.expr.id < 10,
self.expr3.name == 'name1', self.expr.id > 5
])
expected = 'SELECT t2.`name`, t2.`id`, t2.`fid` AS `fid_x`, t2.`isMale`, ' \
't2.`scale`, t2.`birth`, t2.`ds`, t4.`fid` AS `fid_y`, t4.`part1`, t4.`part2` \n' \
'FROM (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table` t1 \n' \
' WHERE (t1.`id` < 10) AND (t1.`id` > 5)\n' \
') t2 \n' \
'INNER JOIN \n' \
' (\n' \
' SELECT * \n' \
' FROM mocked_project.`pyodps_test_expr_table2` t3 \n' \
' WHERE t3.`name` == \'name1\'\n' \
' ) t4\n' \
'ON (t2.`name` == t4.`name`) AND (t2.`id` == t4.`id`)'
self.assertEqual(
to_str(expected),
to_str(ODPSEngine(self.odps).compile(expr, prettify=False)))
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(['name', 'id', 'fid', 'isMale', 'scale', 'birth'],
datatypes('string', 'int64', 'float64', 'boolean', 'decimal', 'datetime'))
self.schema = df_schema_to_odps_schema(schema)
import pandas as pd
self.df = pd.DataFrame(None, columns=schema.names)
self.expr = CollectionExpr(_source_data=self.df, _schema=schema)
self.engine = PandasEngine(self.odps)
self.odps_engine = ODPSEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def _gen_data(self, rows=None, data=None, nullable_field=None, value_range=None):
if data is None:
data = []
for _ in range(rows):
record = []
for t in self.schema.types:
method = getattr(self, '_gen_random_%s' % t.name)
if t.name == 'bigint':
record.append(method(value_range=value_range))
else:
record.append(method())
data.append(record)
if nullable_field is not None:
j = self.schema._name_indexes[nullable_field]
for i, l in enumerate(data):
if i % 2 == 0:
data[i][j] = None
import pandas as pd
self.expr._source_data = pd.DataFrame(data, columns=self.schema.names)
return data
def testBase(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr[::2]
result = self._get_result(self.engine.execute(expr).values)
self.assertEqual(data[::2], result)
expr = self.expr[self.expr.id < 10]['name', lambda x: x.id]
result = self._get_result(self.engine.execute(expr).values)
self.assertEqual(len([it for it in data if it[1] < 10]), len(result))
if len(result) > 0:
self.assertEqual(2, len(result[0]))
expr = self.expr[Scalar(3).rename('const'), self.expr.id, (self.expr.id + 1).rename('id2')]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual([c.name for c in res.columns], ['const', 'id', 'id2'])
self.assertTrue(all(it[0] == 3 for it in result))
self.assertEqual(len(data), len(result))
self.assertEqual([it[1]+1 for it in data], [it[2] for it in result])
expr = self.expr.sort('id')[1:5:2]
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual(sorted(data, key=lambda it: it[1])[1:5:2], result)
res = self.expr.head(10)
result = self._get_result(res.values)
self.assertEqual(data[:10], result)
expr = self.expr.name.hash()
res = self.engine.execute(expr)
result = self._get_result(res.values)
self.assertEqual([[hash(r[0])] for r in data], result),
expr = self.expr.sample(parts=10)
res = self.engine.execute(expr)
self.assertGreaterEqual(len(res), 1)
def testElement(self):
data = self._gen_data(5, nullable_field='name')
fields = [
self.expr.name.isnull().rename('name1'),
self.expr.name.notnull().rename('name2'),
self.expr.name.fillna('test').rename('name3'),
self.expr.id.isin([1, 2, 3]).rename('id1'),
self.expr.id.isin(self.expr.fid.astype('int')).rename('id2'),
self.expr.id.notin([1, 2, 3]).rename('id3'),
self.expr.id.notin(self.expr.fid.astype('int')).rename('id4'),
self.expr.id.between(self.expr.fid, 3).rename('id5'),
self.expr.name.fillna('test').switch('test', 'test' + self.expr.name.fillna('test'),
'test2', 'test2' + self.expr.name.fillna('test'),
default=self.expr.name).rename('name4'),
self.expr.id.cut([100, 200, 300],
labels=['xsmall', 'small', 'large', 'xlarge'],
include_under=True, include_over=True).rename('id6')
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual(len([it for it in data if it[0] is None]),
len([it[0] for it in result if it[0]]))
self.assertEqual(len([it[0] for it in data if it[0] is not None]),
len([it[1] for it in result if it[1]]))
self.assertEqual([(it[0] if it[0] is not None else 'test') for it in data],
[it[2] for it in result])
self.assertEqual([(it[1] in (1, 2, 3)) for it in data],
[it[3] for it in result])
fids = [int(it[2]) for it in data]
self.assertEqual([(it[1] in fids) for it in data],
[it[4] for it in result])
self.assertEqual([(it[1] not in (1, 2, 3)) for it in data],
[it[5] for it in result])
self.assertEqual([(it[1] not in fids) for it in data],
[it[6] for it in result])
self.assertEqual([(it[2] <= it[1] <= 3) for it in data],
[it[7] for it in result])
self.assertEqual([to_str('testtest' if it[0] is None else it[0]) for it in data],
[to_str(it[8]) for it in result])
def get_val(val):
if val <= 100:
return 'xsmall'
elif 100 < val <= 200:
return 'small'
elif 200 < val <= 300:
return 'large'
else:
return 'xlarge'
self.assertEqual([to_str(get_val(it[1])) for it in data], [to_str(it[9]) for it in result])
def testArithmetic(self):
data = self._gen_data(5, value_range=(-1000, 1000))
fields = [
(self.expr.id + 1).rename('id1'),
(self.expr.fid - 1).rename('fid1'),
(self.expr.scale * 2).rename('scale1'),
(self.expr.scale + self.expr.id).rename('scale2'),
(self.expr.id / 2).rename('id2'),
(self.expr.id ** -2).rename('id3'),
abs(self.expr.id).rename('id4'),
(~self.expr.id).rename('id5'),
(-self.expr.fid).rename('fid2'),
(~self.expr.isMale).rename('isMale1'),
(-self.expr.isMale).rename('isMale2'),
(self.expr.id // 2).rename('id6'),
(self.expr.birth + day(1).rename('birth1')),
(self.expr.birth - (self.expr.birth - millisecond(10))).rename('birth2'),
]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(data), len(result))
self.assertEqual([it[1] + 1 for it in data],
[it[0] for it in result])
self.assertAlmostEqual([it[2] - 1 for it in data],
[it[1] for it in result])
self.assertEqual([it[4] * 2 for it in data],
[it[2] for it in result])
self.assertEqual([it[4] + it[1] for it in data],
[it[3] for it in result])
self.assertAlmostEqual([float(it[1]) / 2 for it in data],
[it[4] for it in result])
self.assertEqual([int(it[1] ** -2) for it in data],
[it[5] for it in result])
self.assertEqual([abs(it[1]) for it in data],
[it[6] for it in result])
self.assertEqual([~it[1] for it in data],
[it[7] for it in result])
self.assertAlmostEqual([-it[2] for it in data],
[it[8] for it in result])
self.assertEqual([not it[3] for it in data],
[it[9] for it in result])
self.assertEqual([it[1] // 2 for it in data],
[it[11] for it in result])
self.assertEqual([it[5] + timedelta(days=1) for it in data],
[it[12] for it in result])
self.assertEqual([10] * len(data), [it[13] for it in result])
def testMath(self):
data = self._gen_data(5, value_range=(1, 90))
import numpy as np
methods_to_fields = [
(np.sin, self.expr.id.sin()),
(np.cos, self.expr.id.cos()),
(np.tan, self.expr.id.tan()),
(np.sinh, self.expr.id.sinh()),
(np.cosh, self.expr.id.cosh()),
(np.tanh, self.expr.id.tanh()),
(np.log, self.expr.id.log()),
(np.log2, self.expr.id.log2()),
(np.log10, self.expr.id.log10()),
(np.log1p, self.expr.id.log1p()),
(np.exp, self.expr.id.exp()),
(np.expm1, self.expr.id.expm1()),
(np.arccosh, self.expr.id.arccosh()),
(np.arcsinh, self.expr.id.arcsinh()),
(np.arctanh, self.expr.id.arctanh()),
(np.arctan, self.expr.id.arctan()),
(np.sqrt, self.expr.id.sqrt()),
(np.abs, self.expr.id.abs()),
(np.ceil, self.expr.id.ceil()),
(np.floor, self.expr.id.floor()),
(np.trunc, self.expr.id.trunc()),
]
fields = [it[1].rename('id'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[1]) for it in data]
second = [it[i] for it in result]
self.assertEqual(len(first), len(second))
for it1, it2 in zip(first, second):
if isinstance(it1, float) and np.isnan(it1) and it2 is None:
continue
self.assertAlmostEqual(it1, it2)
def testString(self):
data = self._gen_data(5)
methods_to_fields = [
(lambda s: s.capitalize(), self.expr.name.capitalize()),
(lambda s: data[0][0] in s, self.expr.name.contains(data[0][0], regex=False)),
(lambda s: s.count(data[0][0]), self.expr.name.count(data[0][0])),
(lambda s: s.endswith(data[0][0]), self.expr.name.endswith(data[0][0])),
(lambda s: s.startswith(data[0][0]), self.expr.name.startswith(data[0][0])),
(lambda s: s.find(data[0][0]), self.expr.name.find(data[0][0])),
(lambda s: s.rfind(data[0][0]), self.expr.name.rfind(data[0][0])),
(lambda s: s.replace(data[0][0], 'test'), self.expr.name.replace(data[0][0], 'test')),
(lambda s: s[0], self.expr.name.get(0)),
(lambda s: len(s), self.expr.name.len()),
(lambda s: s.ljust(10), self.expr.name.ljust(10)),
(lambda s: s.ljust(20, '*'), self.expr.name.ljust(20, fillchar='*')),
(lambda s: s.rjust(10), self.expr.name.rjust(10)),
(lambda s: s.rjust(20, '*'), self.expr.name.rjust(20, fillchar='*')),
(lambda s: s * 4, self.expr.name.repeat(4)),
(lambda s: s[2: 10: 2], self.expr.name.slice(2, 10, 2)),
(lambda s: s[-5: -1], self.expr.name.slice(-5, -1)),
(lambda s: s.title(), self.expr.name.title()),
(lambda s: s.rjust(20, '0'), self.expr.name.zfill(20)),
(lambda s: s.isalnum(), self.expr.name.isalnum()),
(lambda s: s.isalpha(), self.expr.name.isalpha()),
(lambda s: s.isdigit(), self.expr.name.isdigit()),
(lambda s: s.isspace(), self.expr.name.isspace()),
(lambda s: s.isupper(), self.expr.name.isupper()),
(lambda s: s.istitle(), self.expr.name.istitle()),
(lambda s: to_str(s).isnumeric(), self.expr.name.isnumeric()),
(lambda s: to_str(s).isdecimal(), self.expr.name.isdecimal()),
]
fields = [it[1].rename('id'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = [method(it[0]) for it in data]
second = [it[i] for it in result]
self.assertEqual(first, second)
def testDatetime(self):
data = self._gen_data(5)
import pandas as pd
methods_to_fields = [
(lambda s: list(s.birth.dt.year.values), self.expr.birth.year),
(lambda s: list(s.birth.dt.month.values), self.expr.birth.month),
(lambda s: list(s.birth.dt.day.values), self.expr.birth.day),
(lambda s: list(s.birth.dt.hour.values), self.expr.birth.hour),
(lambda s: list(s.birth.dt.minute.values), self.expr.birth.minute),
(lambda s: list(s.birth.dt.second.values), self.expr.birth.second),
(lambda s: list(s.birth.dt.weekofyear.values), self.expr.birth.weekofyear),
(lambda s: list(s.birth.dt.dayofweek.values), self.expr.birth.dayofweek),
(lambda s: list(s.birth.dt.weekday.values), self.expr.birth.weekday),
(lambda s: list(s.birth.dt.date.values), self.expr.birth.date),
(lambda s: list(s.birth.dt.strftime('%Y%d')), self.expr.birth.strftime('%Y%d')),
(lambda s: list(s.birth.dt.strftime('%Y%d').map(lambda x: datetime.strptime(x, '%Y%d'))),
self.expr.birth.strftime('%Y%d').strptime('%Y%d')),
]
fields = [it[1].rename('birth'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
second = [it[i] for it in result]
self.assertEqual(first, second)
def testFuncion(self):
data = [
['name1', 4, None, None, None, None],
['name2', 2, None, None, None, None],
['name1', 4, None, None, None, None],
['name1', 3, None, None, None, None],
]
self._gen_data(data=data)
expr = self.expr['id'].map(lambda x: x + 1)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(result, [[r[1] + 1] for r in data])
expr = self.expr['id'].mean().map(lambda x: x + 1)
res = self.engine.execute(expr)
ids = [r[1] for r in data]
self.assertEqual(res, sum(ids) / float(len(ids)) + 1)
expr = self.expr.apply(lambda row: row.name + str(row.id), axis=1, reduce=True).rename('name')
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(result, [[r[0] + str(r[1])] for r in data])
def testFunctionResources(self):
data = self._gen_data(5)
class my_func(object):
def __init__(self, resources):
self.file_resource = resources[0]
self.table_resource = resources[1]
self.valid_ids = [int(l) for l in self.file_resource]
self.valid_ids.extend([int(l[0]) for l in self.table_resource])
def __call__(self, arg):
if isinstance(arg, tuple):
if arg[1] in self.valid_ids:
return arg
else:
if arg in self.valid_ids:
return arg
def my_func2(resources):
file_resource = resources[0]
table_resource = resources[1]
valid_ids = [int(l) for l in file_resource]
valid_ids.extend([int(l[0]) for l in table_resource])
def h(arg):
if isinstance(arg, tuple):
if arg[1] in valid_ids:
return arg
else:
if arg in valid_ids:
return arg
return h
try:
self.odps.delete_resource('pyodps_tmp_file_resource')
except:
pass
file_resource = self.odps.create_resource('pyodps_tmp_file_resource', 'file',
file_obj='\n'.join(str(r[1]) for r in data[:3]))
self.odps.delete_table('pyodps_tmp_table', if_exists=True)
t = self.odps.create_table('pyodps_tmp_table', Schema.from_lists(['id'], ['bigint']))
with t.open_writer() as writer:
writer.write([r[1: 2] for r in data[3: 4]])
try:
self.odps.delete_resource('pyodps_tmp_table_resource')
except:
pass
table_resource = self.odps.create_resource('pyodps_tmp_table_resource', 'table',
table_name=t.name)
try:
expr = self.expr.id.map(my_func, resources=[file_resource, table_resource])
res = self.engine.execute(expr)
result = self._get_result(res)
result = [r for r in result if r[0] is not None]
self.assertEqual(sorted([[r[1]] for r in data[:4]]), sorted(result))
expr = self.expr['name', 'id', 'fid']
expr = expr.apply(my_func, axis=1, resources=[file_resource, table_resource],
names=expr.schema.names, types=expr.schema.types)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(sorted([r[:3] for r in data[:4]]), sorted(result))
expr = self.expr['name', 'id', 'fid']
expr = expr.apply(my_func2, axis=1, resources=[file_resource, table_resource],
names=expr.schema.names, types=expr.schema.types)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(sorted([r[:3] for r in data[:4]]), sorted(result))
finally:
try:
file_resource.drop()
except:
pass
try:
t.drop()
except:
pass
try:
table_resource.drop()
except:
pass
def testApply(self):
data = [
['name1', 4, None, None, None, None],
['name2', 2, None, None, None, None],
['name1', 4, None, None, None, None],
['name1', 3, None, None, None, None],
]
data = self._gen_data(data=data)
def my_func(row):
return row.name
expr = self.expr['name', 'id'].apply(my_func, axis=1, names='name')
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([r[0] for r in result], [r[0] for r in data])
def my_func2(row):
yield len(row.name)
yield row.id
expr = self.expr['name', 'id'].apply(my_func2, axis=1, names='cnt', types='int')
res = self.engine.execute(expr)
result = self._get_result(res)
def gen_expected(data):
for r in data:
yield len(r[0])
yield r[1]
self.assertEqual(sorted([r[0] for r in result]), sorted([r for r in gen_expected(data)]))
def testMapReduceByApplyDistributeSort(self):
data = [
['name key', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key name', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
def mapper(row):
for word in row[0].split():
yield word, 1
class reducer(object):
def __init__(self):
self._curr = None
self._cnt = 0
def __call__(self, row):
if self._curr is None:
self._curr = row.word
elif self._curr != row.word:
yield (self._curr, self._cnt)
self._curr = row.word
self._cnt = 0
self._cnt += row.count
def close(self):
if self._curr is not None:
yield (self._curr, self._cnt)
expr = self.expr['name', ].apply(
mapper, axis=1, names=['word', 'count'], types=['string', 'int'])
expr = expr.groupby('word').sort('word').apply(
reducer, names=['word', 'count'], types=['string', 'int'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
def testMapReduce(self):
data = [
['name key', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key name', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output(['word', 'cnt'], ['string', 'int'])
def mapper(row):
for word in row[0].split():
yield word, 1
@output(['word', 'cnt'], ['string', 'int'])
def reducer(keys):
cnt = [0, ]
def h(row, done):
cnt[0] += row[1]
if done:
yield keys[0], cnt[0]
return h
expr = self.expr['name', ].map_reduce(mapper, reducer, group='word')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
@output(['word', 'cnt'], ['string', 'int'])
class reducer2(object):
def __init__(self, keys):
self.cnt = 0
def __call__(self, row, done):
self.cnt += row.cnt
if done:
yield row.word, self.cnt
expr = self.expr['name', ].map_reduce(mapper, reducer2, group='word')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 3], ['name', 4]]
self.assertEqual(sorted(result), sorted(expected))
def testDistributeSort(self):
data = [
['name', 4, 5.3, None, None, None],
['name', 2, 3.5, None, None, None],
['key', 4, 4.2, None, None, None],
['name', 3, 2.2, None, None, None],
['key', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output_names('name', 'id')
@output_types('string', 'int')
class reducer(object):
def __init__(self):
self._curr = None
self._cnt = 0
def __call__(self, row):
if self._curr is None:
self._curr = row.name
elif self._curr != row.name:
yield (self._curr, self._cnt)
self._curr = row.name
self._cnt = 0
self._cnt += 1
def close(self):
if self._curr is not None:
yield (self._curr, self._cnt)
expr = self.expr['name', ].groupby('name').sort('name').apply(reducer)
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['key', 2], ['name', 3]]
self.assertEqual(sorted(expected), sorted(result))
def testSortDistinct(self):
data = [
['name1', 4, None, None, None, None],
['name2', 2, None, None, None, None],
['name1', 4, None, None, None, None],
['name1', 3, None, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.sort(['name', -self.expr.id]).distinct(['name', lambda x: x.id + 1])[:50]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 3)
expected = [
['name1', 5],
['name1', 4],
['name2', 3]
]
self.assertEqual(expected, result)
def testGroupbyAggregation(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
class Agg(object):
def buffer(self):
return [0]
def __call__(self, buffer, val):
buffer[0] += val
def merge(self, buffer, pbuffer):
buffer[0] += pbuffer[0]
def getvalue(self, buffer):
return buffer[0]
expr = self.expr.groupby(['name', 'id'])[lambda x: x.fid.min() * 2 < 8] \
.agg(self.expr.fid.max() + 1, new_id=self.expr.id.sum(),
new_id2=self.expr.id.agg(Agg))
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 5.1, 6, 6],
['name2', 2, 4.5, 2, 2]
]
result = sorted(result, key=lambda k: k[0])
self.assertEqual(expected, result)
expr = self.expr.groupby(Scalar(1).rename('s')).count()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([5], result[0])
expr = self.expr.groupby(Scalar('const').rename('s')).id.sum()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([16], result[0])
field = self.expr.groupby('name').sort(['id', -self.expr.fid]).row_number()
expr = self.expr['name', 'id', 'fid', field]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 3, 4.1, 1],
['name1', 3, 2.2, 2],
['name1', 4, 5.3, 3],
['name1', 4, 4.2, 4],
['name2', 2, 3.5, 1],
]
result = sorted(result, key=lambda k: (k[0], k[1], -k[2]))
self.assertEqual(expected, result)
expr = self.expr.name.value_counts()[:25]
expected = [
['name1', 4],
['name2', 1]
]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.name.topk(25)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.groupby('name').count()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([it[1:] for it in expected], result)
expected = [
['name1', 2],
['name2', 1]
]
expr = self.expr.groupby('name').id.nunique()
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual([it[1:] for it in expected], result)
expr = self.expr[self.expr['id'] > 2].name.value_counts()[:25]
expected = [
['name1', 4]
]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
expr = self.expr.groupby('name', Scalar(1).rename('constant'))\
.agg(id=self.expr.id.sum())
expected = [
['name1', 1, 14],
['name2', 1, 2]
]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(expected, result)
def testJoinGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.int64, types.int64])
self._gen_data(data=data)
data2 = [
['name1', 4, -1],
['name2', 1, -2]
]
import pandas as pd
expr2 = CollectionExpr(_source_data=pd.DataFrame(data2, columns=schema2.names),
_schema=schema2)
expr = self.expr.join(expr2, on='name')[self.expr]
expr = expr.groupby('id').agg(expr.fid.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
expected = pd.DataFrame(data, columns=self.expr.schema.names).groupby('id').agg({'fid': 'sum'})
self.assertEqual(expected.reset_index().values.tolist(), result)
def testFilterGroupby(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby(['name']).agg(id=self.expr.id.max())[lambda x: x.id > 3]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 1)
expected = [
['name1', 4]
]
self.assertEqual(expected, result)
def testGroupbyProjection(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby('name').agg(id=self.expr.id.max())[
lambda x: 't'+x.name, lambda x: x.id + 1]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['tname1', 5],
['tname2', 3]
]
self.assertEqual(expected, result)
def testWindowFunction(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 6.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr.groupby('name').id.cumsum()
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [[14]] * 4 + [[2]]
self.assertEqual(sorted(expected), sorted(result))
expr = self.expr.groupby('name').sort('fid').id.cummax()
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [[3], [4], [4], [4], [2]]
self.assertEqual(sorted(expected), sorted(result))
expr = self.expr[
self.expr.groupby('name', 'id').sort('fid').id.cummean(),
self.expr.groupby('name').id.cummedian()
]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
[3, 3.5], [3, 3.5], [4, 3.5], [4, 3.5], [2, 2]
]
self.assertEqual(sorted(expected), sorted(result))
expr = self.expr.groupby('name').mutate(id2=lambda x: x.id.cumcount(unique=True),
fid=lambda x: x.fid.cummin(sort='id'))
res = self.engine.execute(expr['name', 'id2', 'fid'])
result = self._get_result(res)
expected = [
['name1', 2, 2.2],
['name1', 2, 2.2],
['name1', 2, 2.2],
['name1', 2, 2.2],
['name2', 1, 3.5],
]
self.assertEqual(sorted(expected), sorted(result))
expr = self.expr[
self.expr.id,
self.expr.groupby('name').rank('id'),
self.expr.groupby('name').dense_rank('fid', ascending=False),
self.expr.groupby('name').row_number(sort=['id', 'fid'], ascending=[True, False]),
self.expr.groupby('name').percent_rank('id'),
]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
[4, 3, 2, 3, float(2) / 3],
[2, 1, 1, 1, 0.0],
[4, 3, 3, 4, float(2) / 3],
[3, 1, 4, 2, float(0) / 3],
[3, 1, 1, 1, float(0) / 3]
]
self.assertEqual(sorted(expected), sorted(result))
expr = self.expr[
self.expr.id,
self.expr.groupby('name').id.lag(offset=3, default=0, sort=['id', 'fid']).rename('id2'),
self.expr.groupby('name').id.lead(offset=1, default=-1,
sort=['id', 'fid'], ascending=[False, False]).rename('id3'),
]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
[4, 3, 4],
[2, 0, -1],
[4, 0, 3],
[3, 0, -1],
[3, 0, 3]
]
self.assertEqual(sorted(expected), sorted(result))
def testWindowRewrite(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
expr = self.expr[self.expr.id - self.expr.id.mean() < 10][
[lambda x: x.id - x.id.max()]][[lambda x: x.id - x.id.min()]][lambda x: x.id - x.id.std() > 0]
res = self.engine.execute(expr)
result = self._get_result(res)
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
expected = df.id - df.id.max()
expected = expected - expected.min()
expected = list(expected[expected - expected.std() > 0])
self.assertEqual(expected, [it[0] for it in result])
def testReduction(self):
data = self._gen_data(rows=5, value_range=(-100, 100))
import pandas as pd
df = pd.DataFrame(data, columns=self.schema.names)
class Agg(object):
def buffer(self):
return [0.0, 0]
def __call__(self, buffer, val):
buffer[0] += val
buffer[1] += 1
def merge(self, buffer, pbuffer):
buffer[0] += pbuffer[0]
buffer[1] += pbuffer[1]
def getvalue(self, buffer):
if buffer[1] == 0:
return 0.0
return buffer[0] / buffer[1]
methods_to_fields = [
(lambda s: df.id.mean(), self.expr.id.mean()),
(lambda s: len(df), self.expr.count()),
(lambda s: df.id.var(ddof=0), self.expr.id.var(ddof=0)),
(lambda s: df.id.std(ddof=0), self.expr.id.std(ddof=0)),
(lambda s: df.id.median(), self.expr.id.median()),
(lambda s: df.id.sum(), self.expr.id.sum()),
(lambda s: df.id.min(), self.expr.id.min()),
(lambda s: df.id.max(), self.expr.id.max()),
(lambda s: df.isMale.min(), self.expr.isMale.min()),
(lambda s: df.name.max(), self.expr.name.max()),
(lambda s: df.birth.max(), self.expr.birth.max()),
(lambda s: df.name.sum(), self.expr.name.sum()),
(lambda s: df.isMale.sum(), self.expr.isMale.sum()),
(lambda s: df.isMale.any(), self.expr.isMale.any()),
(lambda s: df.isMale.all(), self.expr.isMale.all()),
(lambda s: df.name.nunique(), self.expr.name.nunique()),
(lambda s: df.id.mean(), self.expr.id.agg(Agg, rtype='float')),
(lambda s: df.id.count(), self.expr.id.count()),
]
fields = [it[1].rename('f'+str(i)) for i, it in enumerate(methods_to_fields)]
expr = self.expr[fields]
res = self.engine.execute(expr)
result = self._get_result(res)
df = pd.DataFrame(data, columns=self.schema.names)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method(df)
second = [it[i] for it in result][0]
if isinstance(first, float):
self.assertAlmostEqual(first, second)
else:
self.assertEqual(first, second)
self.assertEqual(self.engine.execute(self.expr.id.sum() + 1),
sum(it[1] for it in data) + 1)
expr = self.expr['id', 'fid'].apply(Agg, types=['float'] * 2)
expected = [[df.id.mean()], [df.fid.mean()]]
res = self.engine.execute(expr)
result = self._get_result(res)
for first, second in zip(expected, result):
first = first[0]
second = second[0]
if isinstance(first, float):
self.assertAlmostEqual(first, second)
else:
self.assertEqual(first, second)
def testUserDefinedAggregators(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
@output_types('float')
class Aggregator(object):
def buffer(self):
return [0.0, 0]
def __call__(self, buffer, val):
buffer[0] += val
buffer[1] += 1
def merge(self, buffer, pbuffer):
buffer[0] += pbuffer[0]
buffer[1] += pbuffer[1]
def getvalue(self, buffer):
if buffer[1] == 0:
return 0.0
return buffer[0] / buffer[1]
expr = self.expr.id.agg(Aggregator)
expected = float(16) / 5
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertAlmostEqual(expected, result)
expr = self.expr.groupby(Scalar('const').rename('s')).id.agg(Aggregator)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertAlmostEqual(expected, result[0][0])
expr = self.expr.groupby('name').agg(self.expr.id.agg(Aggregator))
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', float(14)/4],
['name2', 2]
]
for expect_r, actual_r in zip(expected, result):
self.assertEqual(expect_r[0], actual_r[0])
self.assertAlmostEqual(expect_r[1], actual_r[1])
def testJoin(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.int64, types.int64])
self._gen_data(data=data)
data2 = [
['name1', 4, -1],
['name2', 1, -2]
]
import pandas as pd
expr2 = CollectionExpr(_source_data=pd.DataFrame(data2, columns=schema2.names),
_schema=schema2)
expr = self.expr.join(expr2)['name', 'id2']
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 5)
expected = [
[to_str('name1'), 4],
[to_str('name2'), 1]
]
self.assertTrue(all(it in expected for it in result))
expr = self.expr.join(expr2, on=['name', ('id', 'id2')])[self.expr.name, expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(len(result), 2)
expected = [to_str('name1'), 4]
self.assertTrue(all(it == expected for it in result))
expr = self.expr.left_join(expr2, on=['name', ('id', 'id2')])[self.expr.name, expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 4],
['name2', None],
['name1', 4],
['name1', None],
['name1', None]
]
self.assertEqual(len(result), 5)
self.assertTrue(all(it in expected for it in result))
expr = self.expr.right_join(expr2, on=['name', ('id', 'id2')])[self.expr.name, expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 4],
['name1', 4],
[None, 1],
]
self.assertEqual(len(result), 3)
self.assertTrue(all(it in expected for it in result))
expr = self.expr.outer_join(expr2, on=['name', ('id', 'id2')])[self.expr.name, expr2.id2]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 4],
['name1', 4],
['name2', None],
['name1', None],
['name1', None],
[None, 1],
]
self.assertEqual(len(result), 6)
self.assertTrue(all(it in expected for it in result))
def testUnion(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.int64, types.int64])
self._gen_data(data=data)
data2 = [
['name3', 5, -1],
['name4', 6, -2]
]
import pandas as pd
expr2 = CollectionExpr(_source_data=pd.DataFrame(data2, columns=schema2.names),
_schema=schema2)
expr = self.expr['name', 'id'].distinct().union(expr2[expr2.id2.rename('id'), 'name'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [
['name1', 4],
['name1', 3],
['name2', 2],
['name3', 5],
['name4', 6]
]
result = sorted(result)
expected = sorted(expected)
self.assertEqual(len(result), len(expected))
for e, r in zip(result, expected):
self.assertEqual([to_str(t) for t in e],
[to_str(t) for t in r])
def testHllc(self):
names = [randint(0, 100000) for _ in xrange(100000)]
data = [[n] + [None] * 5 for n in names]
self._gen_data(data=data)
expr = self.expr.name.hll_count()
res = self.engine.execute(expr)
result = self._get_result(res)
expect = len(set(names))
self.assertAlmostEqual(expect, result, delta=result*0.1)
def testBloomFilter(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
data2 = [
['name1'],
['name3']
]
self._gen_data(data=data)
schema2 = Schema.from_lists(['name', ], [types.string])
import pandas as pd
expr2 = CollectionExpr(_source_data=pd.DataFrame(data2, columns=schema2.names),
_schema=schema2)
expr = self.expr.bloom_filter('name', expr2[:1].name, capacity=10)
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertTrue(all(r[0] != 'name2' for r in result))
def testPersist(self):
data = [
['name1', 4, 5.3, None, None, None],
['name2', 2, 3.5, None, None, None],
['name1', 4, 4.2, None, None, None],
['name1', 3, 2.2, None, None, None],
['name1', 3, 4.1, None, None, None],
]
self._gen_data(data=data)
table_name = tn('pyodps_test_engine_persist_table')
try:
df = self.engine.persist(self.expr, table_name)
res = df.to_pandas()
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, result)
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
schema = Schema.from_lists(self.schema.names, self.schema.types, ['ds'], ['string'])
self.odps.create_table(table_name, schema)
df = self.engine.persist(self.expr, table_name, partition='ds=today', create_partition=True)
res = self.odps_engine.execute(df)
result = self._get_result(res)
self.assertEqual(len(result), 5)
self.assertEqual(data, [d[:-1] for d in result])
finally:
self.odps.delete_table(table_name, if_exists=True)
try:
self.engine.persist(self.expr, table_name, partitions=['name'])
t = self.odps.get_table(table_name)
self.assertEqual(2, len(list(t.partitions)))
with t.open_reader(partition='name=name1', reopen=True) as r:
self.assertEqual(4, r.count)
with t.open_reader(partition='name=name2', reopen=True) as r:
self.assertEqual(1, r.count)
finally:
self.odps.delete_table(table_name, if_exists=True)
