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', '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'],
datatypes('string', 'int64'))
table = MockTable(name='pyodps_test_expr_table', schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
schema2 = Schema.from_lists(['name2', 'id2'],
datatypes('string', 'int64'))
table2 = MockTable(name='pyodps_test_expr_table2', schema=schema2)
self.expr2 = CollectionExpr(_source_data=table2, _schema=schema2)
def _lines_eq(self, expected, actual):
self.assertSequenceEqual(
[to_str(line.rstrip()) for line in expected.split('\n')],
[to_str(line.rstrip()) for line in actual.split('\n')])
def testProjectionFormatter(self):
expr = self.expr['name', self.expr.id.rename('new_id')].new_id.astype(
'float32')
self._lines_eq(EXPECTED_PROJECTION_FORMAT, repr(expr))
def testFilterFormatter(self):
expr = self.expr[(self.expr.name != 'test') & (self.expr.id > 100)]
self._lines_eq(EXPECTED_FILTER_FORMAT, repr(expr))
def testSliceFormatter(self):
expr = self.expr[:100]
self._lines_eq(EXPECTED_SLICE_FORMAT, repr(expr))
expr = self.expr[5:100:3]
self._lines_eq(EXPECTED_SLICE_WITH_START_STEP_FORMAT, repr(expr))
def testArithmeticFormatter(self):
expr = self.expr
d = -(expr['id']) + 20.34 - expr['id'] + float(20) * expr['id'] \
- expr['id'] / 4.9 + 40 // 2 + expr['id'] // 1.2
try:
self._lines_eq(EXPECTED_ARITHMETIC_FORMAT, repr(d))
except AssertionError as e:
left = [
to_str(line.rstrip())
for line in EXPECTED_ARITHMETIC_FORMAT.split('\n')
]
right = [to_str(line.rstrip()) for line in repr(d).split('\n')]
self.assertEqual(len(left), len(right))
for l, r in zip(left, right):
try:
self.assertEqual(l, r)
except AssertionError:
try:
self.assertAlmostEqual(float(l), float(r))
except:
raise e
def testSortFormatter(self):
expr = self.expr.sort(['name', -self.expr.id])
self._lines_eq(EXPECTED_SORT_FORMAT, repr(expr))
def testDistinctFormatter(self):
expr = self.expr.distinct(['name', self.expr.id + 1])
self._lines_eq(EXPECTED_DISTINCT_FORMAT, repr(expr))
def testGroupbyFormatter(self):
expr = self.expr.groupby(['name', 'id']).agg(new_id=self.expr.id.sum())
self._lines_eq(EXPECTED_GROUPBY_FORMAT, repr(expr))
grouped = self.expr.groupby(['name'])
expr = grouped.mutate(grouped.row_number(sort='id'))
self._lines_eq(EXPECTED_MUTATE_FORMAT, repr(expr))
expr = self.expr.groupby(['name', 'id']).count()
self._lines_eq(EXPECTED_GROUPBY_COUNT_FORMAT, repr(expr))
def testReductionFormatter(self):
expr = self.expr.groupby(['id']).id.std()
self._lines_eq(EXPECTED_REDUCTION_FORMAT, repr(expr))
expr = self.expr.id.mean()
self._lines_eq(EXPECTED_REDUCTION_FORMAT2, repr(expr))
expr = self.expr.count()
self._lines_eq(EXPECTED_REDUCTION_FORMAT3, repr(expr))
def testWindowFormatter(self):
expr = self.expr.groupby(['name']).sort(-self.expr.id).name.rank()
self._lines_eq(EXPECTED_WINDOW_FORMAT1, repr(expr))
expr = self.expr.groupby(['id']).id.cummean(preceding=10,
following=5,
unique=True)
self._lines_eq(EXPECTED_WINDOW_FORMAT2, repr(expr))
def testElementFormatter(self):
expr = self.expr.name.contains('test')
self._lines_eq(EXPECTED_STRING_FORMAT, repr(expr))
expr = self.expr.id.between(1, 3)
self._lines_eq(EXPECTED_ELEMENT_FORMAT, repr(expr))
expr = self.expr.name.astype('datetime').strftime('%Y')
self._lines_eq(EXPECTED_DATETIME_FORMAT, repr(expr))
expr = self.expr.id.switch(3,
self.expr.name,
4,
self.expr.name + 'abc',
default=self.expr.name + 'test')
self._lines_eq(EXPECTED_SWITCH_FORMAT, repr(expr))
def testJoinFormatter(self):
expr = self.expr.join(self.expr2, ('name', 'name2'))
self._lines_eq(EXPECTED_JOIN_FORMAT, repr(expr))
def testAstypeFormatter(self):
expr = self.expr.id.astype('float')
self._lines_eq(EXPECTED_CAST_FORMAT, repr(expr))
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', 'birth', 'scale'][:5],
datatypes('string', 'int64', 'float64', 'boolean', 'datetime',
'decimal')[:5])
self.schema = df_schema_to_odps_schema(schema)
table_name = tn('pyodps_test_%s' % str(uuid.uuid4()).replace('-', '_'))
self.odps.delete_table(table_name, if_exists=True)
self.table = self.odps.create_table(name=table_name,
schema=self.schema)
self.expr = CollectionExpr(_source_data=self.table, _schema=schema)
self.engine = SeahawksEngine(self.odps)
class FakeBar(object):
def update(self, *args, **kwargs):
pass
def inc(self, *args, **kwargs):
pass
def status(self, *args, **kwargs):
pass
self.faked_bar = FakeBar()
def teardown(self):
self.table.drop()
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, data)
return data
def testAsync(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr.id.sum()
future = self.engine.execute(expr, async=True)
self.assertFalse(future.done())
res = future.result()
self.assertEqual(sum(it[1] for it in data), res)
def testCache(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr[self.expr.id < 10].cache()
cnt = expr.count()
dag = self.engine.compile(expr)
self.assertEqual(len(dag.nodes()), 2)
res = self.engine.execute(cnt)
self.assertEqual(len([it for it in data if it[1] < 10]), res)
self.assertTrue(context.is_cached(expr))
table = context.get_cached(expr)
self.assertIsInstance(table, SeahawksTable)
def testBatch(self):
data = self._gen_data(10, value_range=(-1000, 1000))
expr = self.expr[self.expr.id < 10].cache()
expr1 = expr.id.sum()
expr2 = expr.id.mean()
dag = self.engine.compile([expr1, expr2])
self.assertEqual(len(dag.nodes()), 3)
self.assertEqual(sum(len(v) for v in dag._graph.values()), 2)
expect1 = sum(d[1] for d in data if d[1] < 10)
length = len([d[1] for d in data if d[1] < 10])
expect2 = (expect1 / float(length)) if length > 0 else 0.0
res = self.engine.execute([expr1, expr2], n_parallel=2)
self.assertEqual(res[0], expect1)
self.assertAlmostEqual(res[1], expect2)
self.assertTrue(context.is_cached(expr))
# test async and timeout
expr = self.expr[self.expr.id < 10]
expr1 = expr.id.sum()
expr2 = expr.id.mean()
fs = self.engine.execute([expr, expr1, expr2],
n_parallel=2,
async=True,
timeout=1)
self.assertEqual(len(fs), 3)
self.assertEqual(fs[1].result(), expect1)
self.assertAlmostEqual(fs[2].result(), expect2)
self.assertTrue(context.is_cached(expr))
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.assertListAlmostEqual(
sorted(data, key=lambda it: it[1])[:5],
[r[:-1] + [r[-1].replace(tzinfo=None)] for r in result],
only_float=False,
delta=.001)
expr = self.expr[:1].filter(lambda x: x.name == data[1][0])
res = self.engine.execute(expr)
self.assertEqual(len(res), 0)
def testChinese(self):
data = [
['中文', 4, 5.3, None, None],
['\'中文2', 2, 3.5, None, None],
]
self._gen_data(data=data)
expr = self.expr.filter(self.expr.name == '中文')
res = self.engine.execute(expr)
self.assertEqual(len(res), 1)
expr = self.expr.filter(self.expr.name == '\'中文2')
res = self.engine.execute(expr)
self.assertEqual(len(res), 1)
expr = self.expr.filter(self.expr.name == u'中文')
res = self.engine.execute(expr)
self.assertEqual(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.name.fillna('test').switch('test', 1, 'test2',
2).rename('name5'),
self.expr.id.cut([100, 200, 300],
labels=['xsmall', 'small', 'large', 'xlarge'],
include_under=True,
include_over=True).rename('id6'),
self.expr.id.between(self.expr.fid, 3,
inclusive=False).rename('id7'),
]
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])
self.assertEqual([to_str(1 if it[0] is None else None) for it in data],
[to_str(it[9]) 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[10]) for it in result])
self.assertEqual([(it[2] < it[1] < 3) for it in data],
[it[11] 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.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.id % 2).rename('id7'),
]
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.assertListAlmostEqual([it[2] - 1 for it in data],
[it[1] for it in result],
delta=.001)
self.assertListAlmostEqual([float(it[1]) / 2 for it in data],
[it[2] for it in result],
delta=.001)
self.assertEqual([int(it[1]**2) for it in data],
[it[3] for it in result])
self.assertEqual([abs(it[1]) for it in data], [it[4] for it in result])
self.assertEqual([~it[1] for it in data], [it[5] for it in result])
self.assertListAlmostEqual([-it[2] for it in data],
[it[6] for it in result],
delta=.001)
self.assertEqual([not it[3] for it in data], [it[7] for it in result])
self.assertEqual([it[1] // 2 for it in data], [it[9] for it in result])
self.assertEqual([it[1] % 2 for it in data], [it[10] for it in result])
def testMath(self):
# TODO: test sinh, cosh..., and acosh, asinh...
data = self._gen_data(5, value_range=(1, 90))
if hasattr(math, 'expm1'):
expm1 = math.expm1
else:
expm1 = lambda x: 2 * math.exp(x / 2.0) * math.sinh(x / 2.0)
methods_to_fields = [
(math.sin, self.expr.id.sin()),
(math.cos, self.expr.id.cos()),
(math.tan, self.expr.id.tan()),
(math.log, self.expr.id.log()),
(lambda v: math.log(v, 2), self.expr.id.log2()),
(math.log10, self.expr.id.log10()),
(math.log1p, self.expr.id.log1p()),
(math.exp, self.expr.id.exp()),
(expm1, self.expr.id.expm1()),
(math.atan, self.expr.id.arctan()),
(math.sqrt, self.expr.id.sqrt()),
(abs, self.expr.id.abs()),
(math.ceil, self.expr.id.ceil()),
(math.floor, self.expr.id.floor()),
(math.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):
mt = it[0]
def method(v):
try:
return mt(v)
except ValueError:
return float('nan')
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):
not_valid = lambda x: \
x is None or (isinstance(x, float) and (math.isnan(x) or math.isinf(x)))
if not_valid(it1) and not_valid(it2):
continue
if isinstance(it1, float) and it1 > 1.0e15:
scale = 0.1**(int(math.log10(it1)) - 15)
self.assertAlmostEqual(it1 * scale, it2 * scale, delta=8)
else:
self.assertAlmostEqual(it1, it2, delta=2)
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[0] + '|' + str(s[1]),
self.expr.name.cat(self.expr.id.astype('string'), sep='|')),
(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.replace(data[0][0], 'test'),
self.expr.name.replace(data[0][0], 'test', regex=False)),
(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[1:], self.expr.name.slice(1)),
(lambda s: s[1:6], self.expr.name.slice(1, 6)),
(lambda s: s.title(), self.expr.name.title()),
(lambda s: s.rjust(20, '0'), self.expr.name.zfill(20)),
]
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]
if i != 2:
first = [method(it[0]) for it in data]
else:
# cat
first = [method(it) for it in data]
second = [it[i] for it in result]
self.assertEqual(first, second)
def testDatetime(self):
data = self._gen_data(5)
def date_value(sel):
if isinstance(sel, six.string_types):
fun = lambda v: getattr(v, sel)
else:
fun = sel
col_id = [
idx for idx, col in enumerate(self.schema.names)
if col == 'birth'
][0]
return [fun(row[col_id]) for row in data]
methods_to_fields = [
(partial(date_value, 'year'), self.expr.birth.year),
(partial(date_value, 'month'), self.expr.birth.month),
(partial(date_value, 'day'), self.expr.birth.day),
(partial(date_value, 'hour'), self.expr.birth.hour),
(partial(date_value, 'minute'), self.expr.birth.minute),
(partial(date_value, 'second'), self.expr.birth.second),
(partial(date_value, lambda d: d.isocalendar()[1]),
self.expr.birth.weekofyear),
(partial(date_value,
lambda d: d.weekday()), self.expr.birth.dayofweek),
(partial(date_value,
lambda d: d.weekday()), self.expr.birth.weekday),
(partial(date_value, lambda d: time.mktime(d.timetuple())),
self.expr.birth.unix_timestamp),
(partial(
date_value,
lambda d: datetime.combine(d.date(), datetime.min.time())),
self.expr.birth.date),
]
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)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method()
try:
import pandas as pd
def conv(v):
if isinstance(v, pd.Timestamp):
v = v.to_datetime()
if isinstance(v, datetime):
return v.replace(tzinfo=None)
return v
except ImportError:
conv = lambda v: v
second = [conv(it[i]) for it in result]
self.assertEqual(first, second)
def testSortDistinct(self):
data = [
['name1', 4, None, None, None],
['name2', 2, None, None, None],
['name1', 4, None, None, None],
['name1', 3, 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(sorted(expected), sorted(result))
def testPivotTable(self):
data = [['name1', 1, 1.0, True, None], ['name1', 1, 5.0, True, None],
['name1', 2, 2.0, True, None], ['name2', 1, 3.0, False, None],
['name2', 3, 4.0, False, None]]
self._gen_data(data=data)
expr = self.expr
expr1 = expr.pivot_table(rows='name', values='fid')
res = self.engine.execute(expr1)
result = self._get_result(res)
expected = [
['name1', 8.0 / 3],
['name2', 3.5],
]
self.assertListAlmostEqual(sorted(result),
sorted(expected),
only_float=False)
expr2 = expr.pivot_table(rows='name',
values='fid',
aggfunc=['mean', 'sum'])
res = self.engine.execute(expr2)
result = self._get_result(res)
expected = [
['name1', 8.0 / 3, 8.0],
['name2', 3.5, 7.0],
]
self.assertEqual(res.schema.names, ['name', 'fid_mean', 'fid_sum'])
self.assertListAlmostEqual(sorted(result),
sorted(expected),
only_float=False)
expr5 = expr.pivot_table(rows='id',
values='fid',
columns='name',
aggfunc=['mean', 'sum'])
expr6 = expr5['name1_fid_mean',
expr5.groupby(Scalar(1)).sort('name1_fid_mean').
name1_fid_mean.astype('float').cumsum()]
k = lambda x: list(0 if it is None else it for it in x)
expected = [[2, 2], [3, 5], [None, 5]]
res = self.engine.execute(expr6)
result = self._get_result(res)
self.assertEqual(sorted(result, key=k), sorted(expected, key=k))
expr3 = expr.pivot_table(rows='id',
values='fid',
columns='name',
fill_value=0).distinct()
res = self.engine.execute(expr3)
result = self._get_result(res)
expected = [
[2, 0, 2.0],
[3, 4.0, 0],
[1, 3.0, 3.0],
]
self.assertEqual(res.schema.names,
['id', 'name2_fid_mean', 'name1_fid_mean'])
self.assertEqual(result, expected)
expr7 = expr.pivot_table(rows='id',
values='fid',
columns='name',
aggfunc=['mean', 'sum']).cache()
self.assertEqual(len(self.engine.execute(expr7)), 3)
expr8 = self.expr.pivot_table(rows='id', values='fid', columns='name')
self.assertEqual(len(self.engine.execute(expr8)), 3)
self.assertNotIsInstance(expr8.schema, DynamicSchema)
expr9 = (expr8['name1_fid_mean'] -
expr8['name2_fid_mean']).rename('substract')
self.assertEqual(len(self.engine.execute(expr9)), 3)
expr10 = expr8.distinct()
self.assertEqual(len(self.engine.execute(expr10)), 3)
def testGroupbyAggregation(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
self._gen_data(data=data)
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(dropna=True)[: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(sorted([it[1:] for it in expected]), sorted(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)
expr = self.expr[:1]
expr = expr.groupby('name').agg(expr.id.sum())
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['name1', 4]]
self.assertEqual(expected, result)
def testProjectionGroupbyFilter(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
self._gen_data(data=data)
df = self.expr.copy()
df['id'] = df.id + 1
df2 = df.groupby('name').agg(
id=df.id.sum())[lambda x: x.name == 'name2']
expected = [['name2', 3]]
res = self.engine.execute(df2)
result = self._get_result(res)
self.assertEqual(expected, result)
def testJoinGroupby(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = tn('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, 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)
id_idx = [
idx for idx, col in enumerate(self.expr.schema.names)
if col == 'id'
][0]
fid_idx = [
idx for idx, col in enumerate(self.expr.schema.names)
if col == 'fid'
][0]
expected = [[k, sum(
v[fid_idx] for v in row)] for k, row in itertools.groupby(
sorted(data, key=lambda r: r[id_idx]), lambda r: r[id_idx])]
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],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, 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 testWindowFunction(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 6.1, 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(), ]
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [[3], [3], [4], [4], [2]]
self.assertEqual(sorted(expected), sorted(result))
expr = self.expr.groupby('name').mutate(
id2=lambda x: x.id.cumcount(),
fid2=lambda x: x.fid.cummin(sort='id'))
res = self.engine.execute(expr['name', 'id2', 'fid2'])
result = self._get_result(res)
expected = [
['name1', 4, 2.2],
['name1', 4, 2.2],
['name1', 4, 2.2],
['name1', 4, 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.assertListAlmostEqual(l, r)
for l, r in zip(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],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, 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)
id_idx = [
idx for idx, col in enumerate(self.expr.schema.names)
if col == 'id'
][0]
expected = [r[id_idx] for r in data]
maxv = max(expected)
expected = [v - maxv for v in expected]
minv = min(expected)
expected = [v - minv for v in expected]
meanv = sum(expected) * 1.0 / len(expected)
meanv2 = sum([v**2 for v in expected]) * 1.0 / len(expected)
std = math.sqrt(meanv2 - meanv**2)
expected = [v for v in expected if v > std]
self.assertEqual(expected, [it[0] for it in result])
def testReduction(self):
data = self._gen_data(rows=5, value_range=(-100, 100))
def stats(col, func):
col_idx = [
idx for idx, cn in enumerate(self.expr.schema.names)
if cn == col
][0]
return func([r[col_idx] for r in data])
def var(vct, ddof=0):
meanv = mean(vct)
meanv2 = mean([v**2 for v in vct])
return (meanv2 - meanv**2) * len(vct) / (len(vct) - ddof)
def moment(vct, order, central=False, absolute=False):
abs_fun = abs if absolute else lambda x: x
if central:
m = mean(vct)
return mean([abs_fun(v - m)**order for v in vct])
else:
return mean([abs_fun(v)**order for v in vct])
def skew(vct):
n = len(vct)
return moment(vct, 3, central=True) / (std(
vct, 1)**3) * (n**2) / (n - 1) / (n - 2)
def kurtosis(vct):
n = len(vct)
m4 = moment(vct, 4, central=True)
m2 = var(vct, 0)
return 1.0 / (n - 2) / (n - 3) * ((n * n - 1.0) * m4 / m2**2 - 3 *
(n - 1)**2)
mean = lambda v: sum(v) * 1.0 / len(v)
std = lambda v, ddof=0: math.sqrt(var(v, ddof))
nunique = lambda v: len(set(v))
cat = lambda v: len([it for it in v if it is not None])
methods_to_fields = [
(partial(stats, 'id', mean), self.expr.id.mean()),
(partial(len, data), self.expr.count()),
(partial(stats, 'id', var), self.expr.id.var(ddof=0)),
(partial(stats, 'id',
lambda x: var(x, 1)), self.expr.id.var(ddof=1)),
(partial(stats, 'id', std), self.expr.id.std(ddof=0)),
(partial(stats, 'id', lambda x: moment(x, 3, central=True)),
self.expr.id.moment(3, central=True)),
(partial(stats, 'id', skew), self.expr.id.skew()),
(partial(stats, 'id', kurtosis), self.expr.id.kurtosis()),
(partial(stats, 'id', sum), self.expr.id.sum()),
(partial(stats, 'id', min), self.expr.id.min()),
(partial(stats, 'id', max), self.expr.id.max()),
(partial(stats, 'isMale', min), self.expr.isMale.min()),
(partial(stats, 'isMale', sum), self.expr.isMale.sum()),
(partial(stats, 'isMale', any), self.expr.isMale.any()),
(partial(stats, 'isMale', all), self.expr.isMale.all()),
(partial(stats, 'name', nunique), self.expr.name.nunique()),
(partial(stats, 'name', cat), self.expr.name.cat(sep='|')),
(partial(stats, 'id', lambda x: len(x)), 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)
for i, it in enumerate(methods_to_fields):
method = it[0]
first = method()
second = [it[i] for it in result][0]
if i == len(methods_to_fields) - 2: # cat
second = len(second.split('|'))
if isinstance(first, float):
self.assertAlmostEqual(first, second)
else:
if first != second:
pass
self.assertEqual(first, second)
def testJoin(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = tn('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, 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))
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))
grouped = self.expr.groupby('name').agg(
new_id=self.expr.id.sum()).cache()
self.engine.execute(self.expr.join(grouped, on='name'))
expr = self.expr.join(expr2, on=[
'name', ('id', 'id2')
])[lambda x: x.groupby(Scalar(1)).sort('name').row_number(), ]
self.engine.execute(expr)
finally:
table2.drop()
def testUnion(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
schema2 = Schema.from_lists(['name', 'id2', 'id3'],
[types.string, types.bigint, types.bigint])
table_name = tn('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, 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 testScaleValue(self):
data = [
['name1', 4, 5.3],
['name2', 2, 3.5],
['name1', 4, 4.2],
['name1', 3, 2.2],
['name1', 3, 4.1],
]
schema = Schema.from_lists(['name', 'id', 'fid'],
[types.string, types.bigint, types.double])
table_name = tn('pyodps_test_engine_scale_table')
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(name=table_name, schema=schema)
self.odps.write_table(table_name, 0, data)
expr_input = CollectionExpr(_source_data=table,
_schema=odps_schema_to_df_schema(schema))
expr = expr_input.min_max_scale(columns=['fid'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['name1', 4, 1.0], ['name2', 2, 0.41935483870967744],
['name1', 4, 0.6451612903225807], ['name1', 3, 0.0],
['name1', 3, 0.6129032258064515]]
result = sorted(result)
expected = sorted(expected)
for first, second in zip(result, expected):
self.assertEqual(len(first), len(second))
for it1, it2 in zip(first, second):
self.assertAlmostEqual(it1, it2)
expr = expr_input.std_scale(columns=['fid'])
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [['name1', 4, 1.4213602653434203],
['name2', 2, -0.3553400663358544],
['name1', 4, 0.3355989515394193],
['name1', 3, -1.6385125281042194],
['name1', 3, 0.23689337755723686]]
result = sorted(result)
expected = sorted(expected)
for first, second in zip(result, expected):
self.assertEqual(len(first), len(second))
for it1, it2 in zip(first, second):
self.assertAlmostEqual(it1, it2)
def testPersist(self):
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
self._gen_data(data=data)
table_name = tn('pyodps_test_engine_persist_seahawks_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, [r[:-1] for r 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 testMakeKV(self):
from odps import types as odps_types
data = [
['name1', 1.0, 3.0, None, 10.0, None, None],
['name1', None, 3.0, 5.1, None, None, None],
['name1', 7.1, None, None, None, 8.2, None],
['name2', None, 1.2, 1.5, None, None, None],
['name2', None, 1.0, None, None, None, 1.1],
]
kv_cols = ['k1', 'k2', 'k3', 'k5', 'k7', 'k9']
schema = Schema.from_lists(['name'] + kv_cols, [odps_types.string] +
[odps_types.double] * 6)
table_name = tn('pyodps_test_engine_make_kv')
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(name=table_name, schema=schema)
expr = CollectionExpr(_source_data=table,
_schema=odps_schema_to_df_schema(schema))
try:
self.odps.write_table(table, 0, data)
expr1 = expr.to_kv(columns=kv_cols, kv_delim='=')
res = self.engine.execute(expr1)
result = self._get_result(res)
expected = [
['name1', 'k1=1,k2=3,k5=10'],
['name1', 'k2=3,k3=5.1'],
['name1', 'k1=7.1,k7=8.2'],
['name2', 'k2=1.2,k3=1.5'],
['name2', 'k2=1,k9=1.1'],
]
self.assertListEqual(result, expected)
finally:
table.drop()
def testFilterOrder(self):
table_name = tn('pyodps_test_division_error')
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(table_name,
'divided bigint, divisor bigint',
lifecycle=1)
try:
self.odps.write_table(table_name,
[[2, 0], [1, 1], [1, 2], [5, 1], [5, 0]])
df = CollectionExpr(_source_data=table,
_schema=odps_schema_to_df_schema(table.schema))
fdf = df[df.divisor > 0]
ddf = fdf[(fdf.divided / fdf.divisor).rename('result'), ]
expr = ddf[ddf.result > 1]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(result, [[
5,
]])
finally:
table.drop()
def testAXFException(self):
import sqlalchemy
data = [
['name1', 4, 5.3, None, None],
['name2', 2, 3.5, None, None],
['name1', 4, 4.2, None, None],
['name1', 3, 2.2, None, None],
['name1', 3, 4.1, None, None],
]
self._gen_data(data=data)
table_name = tn('pyodps_test_engine_axf_seahawks_table')
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)
with self.assertRaises(sqlalchemy.exc.DatabaseError):
self.engine.execute(df.input)
finally:
self.odps.delete_table(table_name, if_exists=True)
class Test(TestBase):
def setup(self):
datatypes = lambda *types: [validate_data_type(t) for t in types]
schema = Schema.from_lists(["name", "id"], datatypes("string", "int64"))
table = MockTable(name="pyodps_test_expr_table", schema=schema)
self.expr = CollectionExpr(_source_data=table, _schema=schema)
schema2 = Schema.from_lists(["name2", "id2"], datatypes("string", "int64"))
table2 = MockTable(name="pyodps_test_expr_table2", schema=schema2)
self.expr2 = CollectionExpr(_source_data=table2, _schema=schema2)
def _lines_eq(self, expected, actual):
self.assertSequenceEqual(
[to_str(line.rstrip()) for line in expected.split("\n")],
[to_str(line.rstrip()) for line in actual.split("\n")],
)
def testProjectionFormatter(self):
expr = self.expr["name", self.expr.id.rename("new_id")].new_id.astype("float32")
self._lines_eq(EXPECTED_PROJECTION_FORMAT, repr(expr))
def testFilterFormatter(self):
expr = self.expr[(self.expr.name != "test") & (self.expr.id > 100)]
self._lines_eq(EXPECTED_FILTER_FORMAT, repr(expr))
def testSliceFormatter(self):
expr = self.expr[:100]
self._lines_eq(EXPECTED_SLICE_FORMAT, repr(expr))
expr = self.expr[5:100:3]
self._lines_eq(EXPECTED_SLICE_WITH_START_STEP_FORMAT, repr(expr))
def testArithmeticFormatter(self):
expr = self.expr
d = -(expr["id"]) + 20.34 - expr["id"] + float(20) * expr["id"] - expr["id"] / 4.9 + 40 // 2 + expr["id"] // 1.2
try:
self._lines_eq(EXPECTED_ARITHMETIC_FORMAT, repr(d))
except AssertionError as e:
left = [to_str(line.rstrip()) for line in EXPECTED_ARITHMETIC_FORMAT.split("\n")]
right = [to_str(line.rstrip()) for line in repr(d).split("\n")]
self.assertEqual(len(left), len(right))
for l, r in zip(left, right):
try:
self.assertEqual(l, r)
except AssertionError:
try:
self.assertAlmostEqual(float(l), float(r))
except:
raise e
def testSortFormatter(self):
expr = self.expr.sort(["name", -self.expr.id])
self._lines_eq(EXPECTED_SORT_FORMAT, repr(expr))
def testDistinctFormatter(self):
expr = self.expr.distinct(["name", self.expr.id + 1])
self._lines_eq(EXPECTED_DISTINCT_FORMAT, repr(expr))
def testGroupbyFormatter(self):
expr = self.expr.groupby(["name", "id"]).agg(new_id=self.expr.id.sum())
self._lines_eq(EXPECTED_GROUPBY_FORMAT, repr(expr))
grouped = self.expr.groupby(["name"])
expr = grouped.mutate(grouped.row_number())
self._lines_eq(EXPECTED_MUTATE_FORMAT, repr(expr))
expr = self.expr.groupby(["name", "id"]).count()
self._lines_eq(EXPECTED_GROUPBY_COUNT_FORMAT, repr(expr))
def testReductionFormatter(self):
expr = self.expr.groupby(["id"]).id.std()
self._lines_eq(EXPECTED_REDUCTION_FORMAT, repr(expr))
expr = self.expr.id.mean()
self._lines_eq(EXPECTED_REDUCTION_FORMAT2, repr(expr))
expr = self.expr.count()
self._lines_eq(EXPECTED_REDUCTION_FORMAT3, repr(expr))
def testWindowFormatter(self):
expr = self.expr.groupby(["name"]).sort(-self.expr.id).name.rank()
self._lines_eq(EXPECTED_WINDOW_FORMAT1, repr(expr))
expr = self.expr.groupby(["id"]).id.cummean(preceding=10, following=5, unique=True)
self._lines_eq(EXPECTED_WINDOW_FORMAT2, repr(expr))
def testElementFormatter(self):
expr = self.expr.name.contains("test")
self._lines_eq(EXPECTED_STRING_FORMAT, repr(expr))
expr = self.expr.id.between(1, 3)
self._lines_eq(EXPECTED_ELEMENT_FORMAT, repr(expr))
expr = self.expr.name.astype("datetime").strftime("%Y")
self._lines_eq(EXPECTED_DATETIME_FORMAT, repr(expr))
expr = self.expr.id.switch(3, self.expr.name, 4, self.expr.name + "abc", default=self.expr.name + "test")
self._lines_eq(EXPECTED_SWITCH_FORMAT, repr(expr))
def testJoinFormatter(self):
expr = self.expr.join(self.expr2, ("name", "name2"))
self._lines_eq(EXPECTED_JOIN_FORMAT, repr(expr))
def testAstypeFormatter(self):
expr = self.expr.id.astype("float")
self._lines_eq(EXPECTED_CAST_FORMAT, repr(expr))
