def persist(self, line):
try:
import pandas as pd
has_pandas = True
except ImportError:
has_pandas = False
self._set_odps()
line = line.strip().strip(';')
frame_name, table_name = line.split(None, 1)
if '.' in table_name:
project_name, table_name = tuple(table_name.split('.', 1))
else:
project_name = None
frame = self.shell.user_ns[frame_name]
if self._odps.exist_table(table_name, project=project_name):
raise TypeError('%s already exists' % table_name)
if isinstance(frame, DataFrame):
frame.persist(name=table_name, project=project_name)
elif has_pandas and isinstance(frame, pd.DataFrame):
frame = DataFrame(frame)
frame.persist(name=table_name, project=project_name)
def persist(self, line):
try:
import pandas as pd
has_pandas = True
except ImportError:
has_pandas = False
self._set_odps()
line = line.strip().strip(';')
frame_name, table_name = line.split(None, 1)
if '.' in table_name:
project_name, table_name = tuple(table_name.split('.', 1))
else:
project_name = None
frame = self.shell.user_ns[frame_name]
if self._odps.exist_table(table_name, project=project_name):
raise TypeError('%s already exists' % table_name)
if isinstance(frame, DataFrame):
frame.persist(name=table_name, project=project_name, notify=False)
elif has_pandas and isinstance(frame, pd.DataFrame):
frame = DataFrame(frame)
frame.persist(name=table_name, project=project_name, notify=False)
html_notify('Persist succeeded')
def testPandasPersistODPS2(self):
import pandas as pd
import numpy as np
data_int8 = np.random.randint(0, 10, (1, ), dtype=np.int8)
data_int16 = np.random.randint(0, 10, (1, ), dtype=np.int16)
data_int32 = np.random.randint(0, 10, (1, ), dtype=np.int32)
data_int64 = np.random.randint(0, 10, (1, ), dtype=np.int64)
data_float32 = np.random.random((1, )).astype(np.float32)
data_float64 = np.random.random((1, )).astype(np.float64)
df = DataFrame(
pd.DataFrame(
OrderedDict([('data_int8', data_int8),
('data_int16', data_int16),
('data_int32', data_int32),
('data_int64', data_int64),
('data_float32', data_float32),
('data_float64', data_float64)])))
tmp_table_name = tn('pyodps_test_mixed_persist_odps2_types')
self.odps.delete_table(tmp_table_name, if_exists=True)
df.persist(tmp_table_name,
lifecycle=1,
drop_table=True,
odps=self.odps)
t = self.odps.get_table(tmp_table_name)
expected_types = [
odps_types.tinyint, odps_types.smallint, odps_types.int_,
odps_types.bigint, odps_types.float_, odps_types.double
]
self.assertEqual(expected_types, t.schema.types)
def testExistingPersist(self):
self.create_iris(IRIS_TABLE)
df = DataFrame(self.odps.get_table(IRIS_TABLE)).append_id()
odps_schema = df_schema_to_odps_schema(df.schema)
cols = list(reversed(odps_schema.columns))
odps_schema = Schema.from_lists([c.name for c in cols], [c.type for c in cols])
self.odps.delete_table(EXISTING_PERSIST_TABLE, if_exists=True)
self.odps.create_table(EXISTING_PERSIST_TABLE, odps_schema)
df.persist(EXISTING_PERSIST_TABLE)
class Test(TestBase):
def setup(self):
import pandas as pd
odps_data = [
['name1', 1],
['name2', 2],
['name1', 3],
]
pd_data = [['name1', 5], ['name2', 6]]
names = ['name', 'id']
types = ['string', 'bigint']
table = tn('pyodps_df_mixed')
self.odps.delete_table(table, if_exists=True)
self.t = self.odps.create_table(table, Schema.from_lists(names, types))
with self.t.open_writer() as w:
w.write([self.t.new_record(r) for r in odps_data])
self.odps_df = DataFrame(self.t)
self.pd_df = DataFrame(pd.DataFrame(pd_data, columns=names))
self.engine = MixedEngine(self.odps)
self.pd_engine = PandasEngine(self.odps)
def teardown(self):
self.t.drop()
def testGroupReduction(self):
expr = self.odps_df.select(self.odps_df,
id2=self.odps_df.id.map(lambda x: x + 1))
expr = expr.groupby('name').id2.sum()
expected = [['name1', 6], ['name2', 3]]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(sorted([[r[1]] for r in expected]), sorted(result))
def assertPandasEqual(self, df1, df2):
from odps.compat import six
from odps import types as o_types
from pandas.util.testing import assert_frame_equal
# compare column types
def get_odps_type(p_type):
for data_type, builtin_type in six.iteritems(
o_types._odps_primitive_to_builtin_types):
if issubclass(p_type.type, builtin_type):
return data_type
types1 = [get_odps_type(dt) for dt in df1.dtypes]
types2 = [get_odps_type(dt) for dt in df2.dtypes]
self.assertSequenceEqual(types1, types2)
assert_frame_equal(df1, df2, check_dtype=False)
def testJoin(self):
expr = self.odps_df.join(self.pd_df, 'name').sort('id_x')
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
def testUnion(self):
expr = self.odps_df.union(self.pd_df).sort(['id', 'name'])
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df.union(self.pd_df).sort(['id', 'name'])).values
self.assertTrue(result.equals(expected))
def testIsIn(self):
expr = self.odps_df['name'].isin(self.pd_df['name']).rename('isin')
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df['name'].isin(
self.pd_df['name']).rename('isin')).values
self.assertTrue(result.equals(expected))
def testMixed(self):
expr = self.odps_df.union(
self.odps_df.join(self.pd_df,
'name')[lambda x: x.name,
lambda x: x.id_x.rename('id')]).sort(
['name', 'id'])
expr = expr[expr['name'].isin(self.pd_df['name'])]
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
test_expr = df.union(
df.join(self.pd_df, 'name')[lambda x: x.name,
lambda x: x.id_x.rename('id')]).sort(
['name', 'id'])
test_expr = test_expr[test_expr['name'].isin(self.pd_df['name'])]
expected = self.pd_engine.execute(test_expr).values
self.assertTrue(result.equals(expected))
def testPandasPersist(self):
import pandas as pd, numpy as np
self.odps.to_global()
tmp_table_name = tn('pyodps_test_mixed_persist')
self.odps.delete_table(tmp_table_name, if_exists=True)
pd_df = pd.DataFrame(np.arange(9).reshape(3, 3), columns=list('abc'))
df = DataFrame(pd_df).persist(tmp_table_name)
self.assertPandasEqual(df.to_pandas(), pd_df)
self.odps.delete_table(tmp_table_name)
def testExecuteCacheTable(self):
df = self.odps_df[self.odps_df.name == 'name1']
result = df.execute().values
self.assertEqual(len(result), 2)
self.assertIsNotNone(df._cache_data)
_, new_df, cbs = self.engine._compile(df)
try:
self.assertIsNotNone(new_df._source_data)
finally:
[cb() for cb in cbs]
df2 = df[:5]
result = df2.execute()
self.assertEqual(len(result), 2)
def testHandleCache(self):
df = self.pd_df['name', self.pd_df.id + 1]
df.execute()
self.assertIsNotNone(df._cache_data)
df2 = df[df.id < 10]
_, new_df2, cbs = self.engine._compile(df2)
try:
self.assertIsNotNone(new_df2.input._source_data)
finally:
[cb() for cb in cbs]
def testCacheTable(self):
df = self.odps_df.join(self.pd_df, 'name').cache()
df2 = df.sort('id_x')
dag = self.engine._compile_dag(df2)
self.assertEqual(len(dag.nodes()), 3)
result = self.engine.execute(df2).values
df3 = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df3.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
self.assertEqual(len(self.engine._generated_table_names), 2)
table = df._cache_data
self.assertEqual(len(df.execute()), len(expected))
self.assertIs(df._cache_data, table)
df4 = df[df.id_x < 3].count()
result = self.engine.execute(df4)
self.assertEqual(result, 2)
self.assertEqual(df4._cache_data, 2)
def testUseCache(self):
df = self.odps_df[self.odps_df['name'] == 'name1']
self.assertEqual(len(df.head(10)), 2)
df._cache_data.drop()
self.assertRaises(ODPSError,
lambda: self.engine.execute(df['name', 'id']))
def plot(**_):
pass
self.assertRaises(ODPSError, lambda: df.plot(x='id', plot_func=plot))
def testPivot(self):
data = [['name1', 1, 1.0, True], ['name1', 2, 2.0, True],
['name2', 1, 3.0, False], ['name2', 3, 4.0, False]]
table_name = tn('pyodps_test_mixed_engine_pivot')
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(
name=table_name,
schema=Schema.from_lists(
['name', 'id', 'fid', 'ismale'],
['string', 'bigint', 'double', 'boolean']))
expr = DataFrame(table)
try:
self.odps.write_table(table, 0, data)
expr1 = expr.pivot(rows='id', columns='name',
values='fid').distinct()
res = self.engine.execute(expr1)
result = self._get_result(res)
expected = [[1, 1.0, 3.0], [2, 2.0, None], [3, None, 4.0]]
self.assertEqual(sorted(result), sorted(expected))
expr2 = expr.pivot(rows='id',
columns='name',
values=['fid', 'ismale'])
res = self.engine.execute(expr2)
result = self._get_result(res)
expected = [[1, 1.0, 3.0, True, False], [2, 2.0, None, True, None],
[3, None, 4.0, None, False]]
self.assertEqual(sorted(result), sorted(expected))
expr3 = expr.pivot(rows='id', columns='name',
values='fid')['name3']
with self.assertRaises(ValueError) as cm:
self.engine.execute(expr3)
self.assertIn('name3', str(cm.exception))
expr4 = expr.pivot(rows='id', columns='name',
values='fid')['id', 'name1']
res = self.engine.execute(expr4)
result = self._get_result(res)
expected = [[1, 1.0], [2, 2.0], [3, None]]
self.assertEqual(sorted(result), sorted(expected))
expr5 = expr.pivot(rows='id', columns='name', values='fid')
expr5 = expr5[expr5, (expr5['name1'].astype('int') +
1).rename('new_name')]
res = self.engine.execute(expr5)
result = self._get_result(res)
expected = [[1, 1.0, 3.0, 2.0], [2, 2.0, None, 3.0],
[3, None, 4.0, None]]
self.assertEqual(sorted(result), sorted(expected))
expr6 = expr.pivot(rows='id', columns='name', values='fid')
expr6 = expr6.join(self.odps_df, on='id')[expr6, 'name']
res = self.engine.execute(expr6)
result = self._get_result(res)
expected = [[1, 1.0, 3.0, 'name1'], [2, 2.0, None, 'name2'],
[3, None, 4.0, 'name1']]
self.assertEqual(sorted(result), sorted(expected))
finally:
table.drop()
def testPivotTable(self):
data = [['name1', 1, 1.0, True], ['name1', 1, 5.0, True],
['name1', 2, 2.0, True], ['name2', 1, 3.0, False],
['name2', 3, 4.0, False]]
table_name = tn('pyodps_test_mixed_engine_pivot_table')
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(
name=table_name,
schema=Schema.from_lists(
['name', 'id', 'fid', 'ismale'],
['string', 'bigint', 'double', 'boolean']))
expr = DataFrame(table)
try:
self.odps.write_table(table, 0, data)
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.assertEqual(sorted(result), sorted(expected))
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.assertEqual(sorted(result), sorted(expected))
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 = [[1, 3.0, 3.0], [2, 2.0, 0], [3, 0, 4.0]]
self.assertEqual(res.schema.names,
['id', 'name1_fid_mean', 'name2_fid_mean'])
self.assertEqual(result, expected)
class Agg(object):
def buffer(self):
return [0]
def __call__(self, buffer, val):
buffer[0] += val
def merge(self, buffer, pbuffer):
buffer[0] += pbuffer[0]
def getvalue(self, buffer):
return buffer[0]
aggfuncs = OrderedDict([('my_sum', Agg), ('mean', 'mean')])
expr4 = expr.pivot_table(rows='id',
values='fid',
columns='name',
fill_value=0,
aggfunc=aggfuncs)
res = self.engine.execute(expr4)
result = self._get_result(res)
expected = [[1, 6.0, 3.0, 3.0, 3.0], [2, 2.0, 0, 2.0, 0],
[3, 0, 4.0, 0, 4.0]]
self.assertEqual(res.schema.names, [
'id', 'name1_fid_my_sum', 'name2_fid_my_sum', 'name1_fid_mean',
'name2_fid_mean'
])
self.assertEqual(result, expected)
finally:
table.drop()
def testHeadAndTail(self):
res = self.odps_df.head(2)
self.assertEqual(len(res), 2)
df = self.odps_df[self.odps_df['name'] == 'name1']
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertIsNotNone(df._cache_data)
res = self.odps_df.tail(2)
self.assertEqual(len(res), 2)
self.assertTrue(all(it > 1 for it in res.values['id']))
self.assertEqual(len(self.odps_df.name.head(2)), 2)
self.assertEqual(len(self.odps_df.name.tail(2)), 2)
res = self.pd_df.head(1)
self.assertEqual(len(res), 1)
df = self.pd_df[self.pd_df['name'] == 'name1']
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertIsNotNone(df._cache_data)
res = self.pd_df.tail(1)
self.assertEqual(len(res), 1)
self.assertEqual(res.values['id'][0], 6)
self.assertEqual(len(self.pd_df.name.head(1)), 1)
self.assertEqual(len(self.pd_df.name.tail(1)), 1)
def testMapReduceWithResource(self):
pd_df2 = self.odps_df.to_pandas(wrap=True)
@output(['name', 'id'], ['string', 'int'])
def reducer(resources):
d = dict()
for r in resources[0]:
if r.name in d:
d[r.name] += r.id
else:
d[r.name] = r.id
def inner(keys):
def h(row, done):
if row.name in d:
d[row.name] += row.id
else:
d[row.name] = row.id
if done:
yield row.name, d[row.name]
return h
return inner
expr = pd_df2.map_reduce(reducer=reducer,
reducer_resources=[self.pd_df],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
odps_df2 = self.pd_df.persist(tn('pyodps_df_mixed2'), odps=self.odps)
try:
expr = self.odps_df.map_reduce(reducer=reducer,
reducer_resources=[odps_df2],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
expr = self.odps_df.map_reduce(reducer=reducer,
reducer_resources=[self.pd_df],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
expr = pd_df2.map_reduce(reducer=reducer,
reducer_resources=[odps_df2],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
finally:
next(odps_df2.data_source()).drop()
def testBloomFilter(self):
import numpy as np
data2 = [['name1'], ['name3']]
table_name = tn('pyodps_test_mixed_engine_bf_table2')
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name,
schema=Schema.from_lists(['name'],
['string']))
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, data2)
try:
expr = self.odps_df.bloom_filter('name',
expr2[:1].name,
capacity=10)
res = self.engine.execute(expr)
self.assertTrue(np.all(res['name'] != 'name2'))
finally:
table2.drop()
def testCachePersist(self):
expr = self.odps_df
data2 = [['name1', 3.2], ['name3', 2.4]]
table_name = tn('pyodps_test_mixed_engine_cp_table2')
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(
name=table_name,
schema=Schema.from_lists(['name', 'fid'], ['string', 'double']))
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, data2)
@output(expr.schema.names, expr.schema.types)
def h(row):
yield row
l = expr.filter(expr.id > 0).apply(h, axis=1).cache()
r = expr2.filter(expr2.fid > 0)
joined = l.join(r, on=['name', r.fid < 4])['id', 'fid'].cache()
output_table = tn('pyodps_test_mixed_engine_cp_output_table')
self.odps.delete_table(output_table, if_exists=True)
schema = Schema.from_lists(['id', 'fid'], ['bigint', 'double'], ['ds'],
['string'])
output_t = self.odps.create_table(output_table,
schema,
if_not_exists=True)
t = joined.persist(output_table,
partition='ds=today',
create_partition=True)
self.assertEqual(len(t.execute()), 2)
output_t.drop()
def testBigintPartitionedCache(self):
table = tn('pyodps_test_bigint_partitioned_cache')
self.odps.delete_table(table, if_exists=True)
expr = self.odps_df.persist(table, partitions=['id'])
@output(['id', 'name'], ['int', 'string'])
def handle(row):
return row.id + 1, row.name
expr = expr['tt' + expr.name, expr.id].cache()
new_expr = expr.map_reduce(mapper=handle)
res = self.engine.execute(new_expr)
self.assertEqual(len(res), 3)
vid_gt_4 = odps.get_table('tl_gt_4_vid_6_6').to_df().to_pandas()['vid']
predict_features = ['sys', 'dia', 'tl', 'hdl', 'ldl']
use_features = [
t for t in train.columns if t != 'vid' and t not in predict_features
and t != 'pos_4' and not 'log' in t
]
pos_eq_1 = test[test['vid'].isin(vid_gt_4)]
test_eq_1 = pos_eq_1.loc[:, use_features]
submission_gt_4 = pos_eq_1.loc[:, ['vid', 'tl']]
train_gt_4 = train[train['tl'] >= 4]
train_gt_4.index = list(range(train_gt_4.shape[0]))
model = GradientBoostingRegressor(learning_rate=0.01,
n_estimators=800,
max_depth=5,
subsample=0.8,
random_state=1,
verbose=1,
min_samples_leaf=20)
gbdt_model(train_gt_4, 'tl', use_features, test_eq_1, submission_gt_4,
model)
gt_4_index = submission[submission['vid'].isin(
submission_gt_4['vid'])].index
submission_temp = submission.loc[gt_4_index, ['vid', 'tl']]
merge_fat = pd.merge(submission_temp, submission_gt_4, on='vid')
temp_columns = [tc for tc in merge_fat.columns if tc != 'vid']
replace_num = np.max(merge_fat.loc[:, temp_columns], axis=1)
submission.loc[gt_4_index, 'tl'] = replace_num.values
print(submission.sort_values(by=['tl'], ascending=False))
sub_final = DataFrame(submission)
sub_final.persist('tl_jz_5_fold_6_6_22_submit_modified_high_value')
class Test(TestBase):
def setup(self):
import pandas as pd
odps_data = [
['name1', 1],
['name2', 2],
['name1', 3],
]
pd_data = [
['name1', 5],
['name2', 6]
]
names = ['name', 'id']
types = ['string', 'bigint']
table = tn('pyodps_df_mixed')
self.odps.delete_table(table, if_exists=True)
self.t = self.odps.create_table(table, Schema.from_lists(names, types))
with self.t.open_writer() as w:
w.write([self.t.new_record(r) for r in odps_data])
self.odps_df = DataFrame(self.t)
self.pd_df = DataFrame(pd.DataFrame(pd_data, columns=names))
self.engine = MixedEngine(self.odps)
self.pd_engine = PandasEngine(self.odps)
def teardown(self):
self.t.drop()
def assertPandasEqual(self, df1, df2):
from odps.compat import six
from odps import types as o_types
from pandas.util.testing import assert_frame_equal
# compare column types
def get_odps_type(p_type):
for data_type, builtin_type in six.iteritems(o_types._odps_primitive_to_builtin_types):
if issubclass(p_type.type, builtin_type):
return data_type
types1 = [get_odps_type(dt) for dt in df1.dtypes]
types2 = [get_odps_type(dt) for dt in df2.dtypes]
self.assertSequenceEqual(types1, types2)
assert_frame_equal(df1, df2, check_dtype=False)
def testJoin(self):
expr = self.odps_df.join(self.pd_df, 'name').sort('id_x')
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
def testUnion(self):
expr = self.odps_df.union(self.pd_df).sort(['id', 'name'])
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df.union(self.pd_df).sort(['id', 'name'])).values
self.assertTrue(result.equals(expected))
def testIsIn(self):
expr = self.odps_df['name'].isin(self.pd_df['name']).rename('isin')
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df['name'].isin(self.pd_df['name']).rename('isin')).values
self.assertTrue(result.equals(expected))
def testMixed(self):
expr = self.odps_df.union(
self.odps_df.join(self.pd_df, 'name')[
lambda x: x.name,
lambda x: x.id_x.rename('id')
]).sort(['name', 'id'])
expr = expr[expr['name'].isin(self.pd_df['name'])]
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
test_expr = df.union(
df.join(self.pd_df, 'name')[
lambda x: x.name,
lambda x: x.id_x.rename('id')
]).sort(['name', 'id'])
test_expr = test_expr[test_expr['name'].isin(self.pd_df['name'])]
expected = self.pd_engine.execute(test_expr).values
self.assertTrue(result.equals(expected))
def testPandasPersist(self):
import pandas as pd, numpy as np
self.odps.to_global()
tmp_table_name = tn('pyodps_test_mixed_persist')
self.odps.delete_table(tmp_table_name, if_exists=True)
pd_df = pd.DataFrame(np.arange(9).reshape(3, 3), columns=list('abc'))
df = DataFrame(pd_df).persist(tmp_table_name)
self.assertPandasEqual(df.to_pandas(), pd_df)
self.odps.delete_table(tmp_table_name)
def testExecuteCacheTable(self):
df = self.odps_df[self.odps_df.name == 'name1']
result = df.execute().values
self.assertEqual(len(result), 2)
self.assertIsNotNone(df._cache_data)
new_df = self.engine._pre_process(df)
_, new_df, cbs = self.engine._compile(new_df)
try:
self.assertIsNotNone(new_df._source_data)
finally:
[cb() for cb in cbs]
df2 = df[:5]
result = df2.execute()
self.assertEqual(len(result), 2)
def testHandleCache(self):
df = self.pd_df['name', self.pd_df.id + 1]
df.execute()
self.assertIsNotNone(df._cache_data)
df2 = df[df.id < 10]
new_df2 = self.engine._pre_process(df2)
_, new_df2, cbs = self.engine._compile(new_df2)
try:
self.assertIsNotNone(new_df2.input._source_data)
finally:
[cb() for cb in cbs]
def testCacheTable(self):
df = self.odps_df.join(self.pd_df, 'name').cache()
df2 = df.sort('id_x')
dag = self.engine._compile_dag(df2)
self.assertEqual(len(dag.nodes()), 3)
result = self.engine.execute(df2).values
df3 = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df3.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
self.assertEqual(len(self.engine._generated_table_names), 2)
table = df._cache_data
self.assertEqual(len(df.execute()), len(expected))
self.assertIs(df._cache_data, table)
df4 = df[df.id_x < 3].count()
result = self.engine.execute(df4)
self.assertEqual(result, 2)
self.assertEqual(df4._cache_data, 2)
def testUseCache(self):
df = self.odps_df[self.odps_df['name'] == 'name1']
self.assertEqual(len(df.head(10)), 2)
df._cache_data.drop()
self.assertRaises(ODPSError, lambda: self.engine.execute(df['name', 'id']))
def plot(**_):
pass
self.assertRaises(ODPSError, lambda: df.plot(x='id', plot_func=plot))
def testHeadAndTail(self):
res = self.odps_df.head(2)
self.assertEqual(len(res), 2)
df = self.odps_df[self.odps_df['name'] == 'name1']
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertIsNotNone(df._cache_data)
res = self.odps_df.tail(2)
self.assertEqual(len(res), 2)
self.assertTrue(all(it > 1 for it in res.values['id']))
self.assertEqual(len(self.odps_df.name.head(2)), 2)
self.assertEqual(len(self.odps_df.name.tail(2)), 2)
res = self.pd_df.head(1)
self.assertEqual(len(res), 1)
df = self.pd_df[self.pd_df['name'] == 'name1']
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertIsNotNone(df._cache_data)
res = self.pd_df.tail(1)
self.assertEqual(len(res), 1)
self.assertEqual(res.values['id'][0], 6)
self.assertEqual(len(self.pd_df.name.head(1)), 1)
self.assertEqual(len(self.pd_df.name.tail(1)), 1)
def testMapReduceWithResource(self):
pd_df2 = self.odps_df.to_pandas(wrap=True)
@output(['name', 'id'], ['string', 'int'])
def reducer(resources):
d = dict()
for r in resources[0]:
if r.name in d:
d[r.name] += r.id
else:
d[r.name] = r.id
def inner(keys):
def h(row, done):
if row.name in d:
d[row.name] += row.id
else:
d[row.name] = row.id
if done:
yield row.name, d[row.name]
return h
return inner
expr = pd_df2.map_reduce(reducer=reducer, reducer_resources=[self.pd_df], group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
odps_df2 = self.pd_df.persist('pyodps_df_mixed2', odps=self.odps)
try:
expr = self.odps_df.map_reduce(reducer=reducer, reducer_resources=[odps_df2], group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
expr = self.odps_df.map_reduce(reducer=reducer, reducer_resources=[self.pd_df], group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
expr = pd_df2.map_reduce(reducer=reducer, reducer_resources=[odps_df2], group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
finally:
next(odps_df2.data_source()).drop()
def testBloomFilter(self):
import numpy as np
data2 = [
['name1'],
['name3']
]
table_name = tn('pyodps_test_mixed_engine_bf_table2')
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name,
schema=Schema.from_lists(['name'], ['string']))
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, [table2.new_record(values=d) for d in data2])
try:
expr = self.odps_df.bloom_filter('name', expr2[:1].name, capacity=10)
res = self.engine.execute(expr)
self.assertTrue(np.all(res['name'] != 'name2'))
finally:
table2.drop()
'juz_train_6_6_add_wzm_for145_final').to_df().to_pandas()
test = odps.get_table(
'juz_test_6_6_add_wzm_for145_final').to_df().to_pandas()
print(train.shape, test.shape)
gene_list = get_one_hot_list(train, use_label, 14)
train.replace(-999, np.nan, inplace=True)
test.replace(-999, np.nan, inplace=True)
drop_snp = [s for s in train.columns if 'snp' in s]
train.drop(drop_snp, axis=1, inplace=True)
test.drop(drop_snp, axis=1, inplace=True)
gene_data = odps.get_table('meinian_round2_snp').to_df().to_pandas()
snp_data = pd.get_dummies(gene_data.loc[:, gene_list])
snp_data['vid'] = gene_data['vid'].values
for s in snp_data.columns:
if s != 'vid':
snp_data[s] = snp_data[s].astype(int)
train_merge = pd.merge(train, snp_data, on='vid', how='left')
test_merge = pd.merge(test, snp_data, on='vid', how='left')
train_merge.fillna(-999, inplace=True)
test_merge.fillna(-999, inplace=True)
print('final train shape:{}, test shape:{} '.format(
train_merge.shape, test_merge.shape))
juz_train = DataFrame(train_merge)
juz_test = DataFrame(test_merge)
juz_train.persist('{}_juz_train_6_6_snp_onehot_22'.format(use_label))
juz_test.persist('{}_juz_test_6_6_snp_onehot_22'.format(use_label))
merge_df = "$".join(list(df['results']))
else:
merge_df = df['results'].values[0]
return merge_df
# 数据简单处理
print(part_1_2.shape)
is_happen = part_1_2.groupby(['vid', 'test_id']).size().reset_index()
# 重塑index用来去重
is_happen['new_index'] = is_happen['vid'] + '_' + is_happen['test_id']
is_happen_new = is_happen[is_happen[0] > 1]['new_index']
part_1_2['new_index'] = part_1_2['vid'] + '_' + part_1_2['test_id']
unique_part = part_1_2[part_1_2['new_index'].isin(list(is_happen_new))]
unique_part = unique_part.sort_values(['vid', 'test_id'])
no_unique_part = part_1_2[~part_1_2['new_index'].isin(list(is_happen_new))]
print('begin')
part_1_2_not_unique = unique_part.groupby(['vid', 'test_id'
]).apply(merge_table).reset_index()
part_1_2_not_unique.rename(columns={0: 'results'}, inplace=True)
tmp = pd.concat(
[part_1_2_not_unique, no_unique_part[['vid', 'test_id', 'results']]])
# 行列转换
print('finish')
tmp = tmp.pivot(index='vid', values='results', columns='test_id')
print(tmp.shape)
combine_data = DataFrame(tmp, unknown_as_string=True)
combine_data.persist('origin_data_combine_part1_part2')
print('total time', time.time() - begin_time)
def gbdt_model(df, label, use_feature, true_test, submission_data, gbdt_model):
print(submission_data.head())
print("基于GBDT： 开始训练 label 为{}...".format(label))
value4preds = df['pos_4']
train_data = df.loc[:, use_feature]
print(train_data.shape, true_test.shape)
pred_labels = np.zeros(df.shape[0])
submission_label = np.zeros((true_test.shape[0], 5))
kf = KFold(n_splits=5, shuffle=True, random_state=1024)
five_fold_index = list(kf.split(train_data, value4preds))
train_index_1, test_index_1 = five_fold_index[0]
print('第1次训练...')
x_train_1, x_test_1 = train_data.iloc[train_index_1], train_data.iloc[test_index_1]
y_train_1, y_test_1 = value4preds.iloc[train_index_1], value4preds.iloc[test_index_1]
gbdt_model.fit(x_train_1, y_train_1)
pred_labels[x_test_1.index] = np.where(gbdt_model.predict(x_test_1) > 0.5, 1, 0)
submission_label[:, 0] = np.where(gbdt_model.predict(true_test) > 0.5, 1, 0)
print('第1次训练结束')
print('*******************************************************************')
train_index_2, test_index_2 = five_fold_index[1]
print('第2次训练...')
x_train_2, x_test_2 = train_data.iloc[train_index_2], train_data.iloc[test_index_2]
y_train_2, y_test_2 = value4preds.iloc[train_index_2], value4preds.iloc[test_index_2]
gbdt_model.fit(x_train_2, y_train_2)
pred_labels[x_test_2.index] = np.where(gbdt_model.predict(x_test_2) > 0.5, 1, 0)
submission_label[:, 1] = np.where(gbdt_model.predict(true_test) > 0.5, 1, 0)
print('第2次训练结束')
print('*******************************************************************')
train_index_3, test_index_3 = five_fold_index[2]
print('第3次训练...')
x_train_3, x_test_3 = train_data.iloc[train_index_3], train_data.iloc[test_index_3]
y_train_3, y_test_3 = value4preds.iloc[train_index_3], value4preds.iloc[test_index_3]
gbdt_model.fit(x_train_3, y_train_3)
pred_labels[x_test_3.index] = np.where(gbdt_model.predict(x_test_3) > 0.5, 1, 0)
submission_label[:, 2] = np.where(gbdt_model.predict(true_test) > 0.5, 1, 0)
print('第3次训练结束')
print('*******************************************************************')
train_index_4, test_index_4 = five_fold_index[3]
print('第4次训练...')
x_train_4, x_test_4 = train_data.iloc[train_index_4], train_data.iloc[test_index_4]
y_train_4, y_test_4 = value4preds.iloc[train_index_4], value4preds.iloc[test_index_4]
gbdt_model.fit(x_train_4, y_train_4)
pred_labels[x_test_4.index] = np.where(gbdt_model.predict(x_test_4) > 0.5, 1, 0)
submission_label[:, 3] = np.where(gbdt_model.predict(true_test) > 0.5, 1, 0)
print('第4次训练结束')
print('*******************************************************************')
train_index_5, test_index_5 = five_fold_index[4]
print('第5次训练...')
x_train_5, x_test_5 = train_data.iloc[train_index_5], train_data.iloc[test_index_5]
y_train_5, y_test_5 = value4preds.iloc[train_index_5], value4preds.iloc[test_index_5]
gbdt_model.fit(x_train_5, y_train_5)
pred_labels[x_test_5.index] = np.where(gbdt_model.predict(x_test_5) > 0.5, 1, 0)
submission_label[:, 4] = np.where(gbdt_model.predict(true_test) > 0.5, 1, 0)
print('第5次训练结束')
print('*******************************************************************')
submission_data['pos_4'] = np.where(np.sum(submission_label, axis=1) >= 1, 1, 0)
print(classification_report(pred_labels, value4preds))
print(submission_data[submission_data['pos_4']==1])
sub_class = DataFrame(submission_data[submission_data['pos_4']==1], unknown_as_string=True)
sub_class.persist('tl_gt_4_vid_6_6')
data_frame['heart_times'] = data_frame['1001'].apply(extract_num_norm)
data_frame['all_result'] = '_'
for p in data_frame.columns:
if p != 'vid':
data_frame['all_result'] = data_frame[
'all_result'] + '_' + data_frame[p].astype('str')
data_frame['gender'] = data_frame['all_result'].apply(is_sex)
del data_frame['all_result']
new_add = [
'xue_ya_pian_gao', 'xue_zhi_pian_gao', 'xue_tang_pian_gao',
'high_sugar', 'guan_xin_bin', 'shen', 'smoke', 'niao', 'heart_rate',
'3399_w', '3301_w', '0403_w', '0421_w', '0405_w', 'gender',
'blood_pipe_style', 'health', 'pres_front', 'pres_back', 'heart_times',
'vid', 'dannan_jieshi', 'dannan_xirou', 'shen_jieshi', 'shen_nanz',
'gan_nanz', 'gan_ying_hua'
]
yy_list.extend(new_add)
return data_frame.loc[:, yy_list]
if __name__ == "__main__":
part_1_2 = odps.get_table(
'origin_data_combine_part1_part2').to_df().to_pandas()
word_data = word2num(part_1_2)
print('the shape of word_data: ', word_data.shape)
juz_word_data = DataFrame(word_data)
juz_word_data.persist('juz_word_data_5_30')
num_ex_str = data_frame.loc[:, total]
return num_ex_str
if __name__ == "__main__":
part_1_2 = odps.get_table(
'origin_data_combine_part1_part2').to_df().to_pandas()
part_1_2['jia_zx'] = split_data(part_1_2['0101'], '甲状腺')
part_1_2['left_shen'] = split_data(part_1_2['0117'], '左肾')
part_1_2['right_shen'] = split_data(part_1_2['0118'], '右肾')
part_1_2_copy = part_1_2.copy(deep=True)
ex_num_data = ex_num_from_str(part_1_2)
print('the shape of the num_data get from word: ', ex_num_data.shape)
pure_num_data = get_pure_num_features(part_1_2_copy, 0.96)
pure_columns = [p for p in pure_num_data.columns if p != 'vid']
ex_num_columns = [
i for i in ex_num_data.columns if i not in
['vid', '314', '1308', '1319', '1320', '1321', '1322', '0424', '0425']
]
print('the shape of origin num data: ', pure_num_data.shape)
numeric_data = pd.merge(pure_num_data, ex_num_data, on='vid', how='inner')
exm_drop = []
for w in pure_columns + ex_num_columns:
if np.abs(numeric_data[w].skew()) > 12:
exm_drop.append(w)
print(exm_drop)
numeric_data.drop(exm_drop, axis=1, inplace=True)
print('total data shape: ', numeric_data.shape)
juz_num_data = DataFrame(numeric_data)
juz_num_data.persist('juz_num_data_5_31')
c2_test = test_merge.loc[(test_merge['2403']!=-999)&(test_merge['2404']==-999)&(test_merge['2405']!=-999), '2404'].index.values
c3_test = test_merge.loc[(test_merge['2403']!=-999)&(test_merge['2404']!=-999)&(test_merge['2405']==-999), '2405'].index.values
for i in c1_test:
test_merge.loc[i, '2403'] = (test_merge.loc[i, '2404']/100) * (test_merge.loc[i, '2404']/100) * (test_merge.loc[i, '2405'])
for i in c2_test:
test_merge.loc[i, '2404'] = np.sqrt((test_merge.loc[i, '2403']) / (test_merge.loc[i, '2405'])) * 100
for i in c3_test:
test_merge.loc[i, '2405'] = (test_merge.loc[i, '2403']) / ((test_merge.loc[i, '2404']/100) * (test_merge.loc[i, '2404']/100))
test_merge.loc[test_merge['2403']==0, '2403'] = -999
test_merge.loc[test_merge['2404']==0, '2404'] = -999
test_merge.loc[test_merge['2405']==0, '2405'] = -999
t_train = DataFrame(train_merge[e_c_cols+c_not_count+['table_count','result_len','result_avg_len','sugar_high_related','sex']])
t_train.persist('wzm_trainset3_3_b2')
t_test = DataFrame(test_merge[e_c_cols+c_not_count+['table_count','result_len','result_avg_len','sugar_high_related','sex']])
t_test.persist('wzm_testset3_3_b2')
'''
exclude_cols = ['vid', 'sys', 'dia', 'tl', 'hdl', 'ldl']
e_c_cols = exclude_cols + cate_cols + cate_cols2
t_train = DataFrame(train_merge[e_c_cols+['sugar_high_related']])
t_train.persist('wzm_trainset3_3_b2')
t_test = DataFrame(test_merge[e_c_cols+['sugar_high_related']])
t_test.persist('wzm_testset3_3_b2')
'jz_xgb_pred_val_1').to_df().to_pandas().loc[:,
['vid', 'log_tl', 'result']]
val_2 = odps.get_table(
'jz_xgb_pred_val_2').to_df().to_pandas().loc[:,
['vid', 'log_tl', 'result']]
val_3 = odps.get_table(
'jz_xgb_pred_val_3').to_df().to_pandas().loc[:,
['vid', 'log_tl', 'result']]
val_4 = odps.get_table(
'jz_xgb_pred_val_4').to_df().to_pandas().loc[:,
['vid', 'log_tl', 'result']]
val_5 = odps.get_table(
'jz_xgb_pred_val_5').to_df().to_pandas().loc[:,
['vid', 'log_tl', 'result']]
xgb_result = val_1.loc[:, ['vid']]
xgb_result['tl'] = np.exp(
(val_1['result'] + val_2['result'] + val_3['result'] + val_4['result'] +
val_5['result']) / 5)
test_odps = DataFrame(xgb_result)
test_odps.persist('tl_xgb_result')
'''
val = pd.concat([val_1, val_2, val_3, val_4, val_5])
print('fold 1: ', eval_metric(np.exp(val_1['result']), np.exp(val_1['log_tl'])))
print('fold 2: ', eval_metric(np.exp(val_2['result']), np.exp(val_2['log_tl'])))
print('fold 3: ', eval_metric(np.exp(val_3['result']), np.exp(val_3['log_tl'])))
print('fold 4: ', eval_metric(np.exp(val_4['result']), np.exp(val_4['log_tl'])))
print('fold 5: ', eval_metric(np.exp(val_5['result']), np.exp(val_5['log_tl'])))
print('total loss: ', eval_metric(np.exp(val['result']), np.exp(val['log_tl'])))
'''
print(gbdt_features)
sfm_rf = SelectFromModel(rf, threshold=0.001)
sfm_rf.fit_transform(x_train, label)
rf_features = set(x_train.columns[sfm_rf.get_support()])
print('*************************************')
print(rf_features)
print(gbdt_features & rf_features)
sfm_l2 = SelectFromModel(l2, threshold=0.5)
sfm_l2.fit_transform(x_train, label)
l2_features = set(x_train.columns[sfm_l2.get_support()])
print('*************************************')
print(l2_features)
final_features = list(gbdt_features | rf_features | l2_features)
# choose top k features
#final_features = list((gbdt_features & rf_features) | l2_features)
print('gbdt model has {} features'.format(len(gbdt_features)))
print('rf model has {} features'.format(len(rf_features)))
print('l2 model has {} features'.format(len(l2_features)))
print('final has {} features'.format(len(final_features)))
print('*************************************')
print(final_features)
print('*************************************')
final_features.extend(['vid', 'tl'])
train_final = DataFrame(train.loc[:, final_features])
train_final.persist('combine_tl_train_6_2')
test_final = DataFrame(test.loc[:, final_features])
test_final.persist('combine_tl_test_6_2')
return eval_metric(scores, np.exp(value4preds))
# b-board
# 739 601 887 709 1221
# 'dia': 0.018069628693683809
if __name__ == "__main__":
train = odps.get_table(
'dia_juz_train_6_6_snp_onehot_22').to_df().to_pandas()
test = odps.get_table('dia_juz_test_6_6_snp_onehot_22').to_df().to_pandas()
print(train.shape)
print(test.shape)
predict_features = ['sys', 'dia', 'tl', 'hdl', 'ldl']
use_features = [
t for t in train.columns if t != 'vid' and t not in predict_features
]
test_data = test.loc[:, use_features]
submission = test.loc[:, ['vid', 'dia']]
base_line_score = np.zeros(5)
start = time.time()
for i, j in enumerate(predict_features):
if j in ['dia']:
base_line_score[i] = gbdt_model(train, j, use_features, test_data,
submission)
print(dict(zip(predict_features, base_line_score)))
print('CV训练用时{}秒'.format(time.time() - start))
print('scores:', np.mean(base_line_score))
sub_final = DataFrame(submission)
sub_final.persist('dia_jz_5_fold_6_6_submit_22')
import numpy as np
import sys
reload(sys)
sys.setdefaultencoding('utf8')
label = 'tl'
train = odps.get_table(
'{}_juz_train_6_6_snp_onehot_22'.format(label)).to_df().to_pandas()
test = odps.get_table(
'{}_juz_test_6_6_snp_onehot_22'.format(label)).to_df().to_pandas()
print(train.shape, test.shape)
train['log_{}'.format(label)] = np.log(train[label])
test['log_{}'.format(label)] = np.log(test[label])
predict_features = ['sys', 'dia', 'tl', 'hdl', 'ldl']
for i in train.columns:
if i != 'vid' and not 'snp' in i and not 'log' in i and i not in predict_features:
train['jz_{}'.format(i)] = train[i]
test['jz_{}'.format(i)] = test[i]
predict_features.append(i)
train.drop(predict_features, axis=1, inplace=True)
test.drop(predict_features, axis=1, inplace=True)
print(train.shape, test.shape)
juz_train = DataFrame(train)
juz_test = DataFrame(test)
juz_train.persist('juz_train_6_7_xgb')
juz_test.persist('juz_test_6_7_xgb')
from odps import ODPS
import pandas as pd
from odps.df import DataFrame
from sklearn.model_selection import KFold
train_data = odps.get_table('juz_train_6_7_xgb').to_df().to_pandas()
kf = KFold(n_splits=5, shuffle=True, random_state=1024)
for t, (train_index, test_index) in enumerate(kf.split(train_data), start=1):
print('第{}次拆分...'.format(t))
x_train, x_test = train_data.iloc[train_index], train_data.iloc[test_index]
print(x_train.shape, x_test.shape)
train_odps = DataFrame(x_train)
test_odps = DataFrame(x_test)
train_odps.persist('tl_xgb_train_{}'.format(t))
test_odps.persist('tl_xgb_test_{}'.format(t))
class Test(TestBase):
def setup(self):
import pandas as pd
odps_data = [
['name1', 1],
['name2', 2],
['name1', 3],
]
pd_data = [['name1', 5], ['name2', 6]]
names = ['name', 'id']
types = ['string', 'bigint']
table = tn('pyodps_df_mixed_%d' % os.getpid())
if self.odps.exist_table(table):
self.t = self.odps.get_table(table)
else:
self.t = self.odps.create_table(table,
Schema.from_lists(names, types),
lifecycle=1)
with self.t.open_writer() as w:
w.write([self.t.new_record(r) for r in odps_data])
self.odps_df = DataFrame(self.t)
self.pd_df = DataFrame(pd.DataFrame(pd_data, columns=names))
self.engine = MixedEngine(self.odps)
self.pd_engine = PandasEngine(self.odps)
def teardown(self):
self.engine._selecter.force_odps = False
def testGroupReduction(self):
expr = self.odps_df.select(self.odps_df,
id2=self.odps_df.id.map(lambda x: x + 1))
expr = expr.groupby('name').id2.sum()
expected = [['name1', 6], ['name2', 3]]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(sorted([[r[1]] for r in expected]), sorted(result))
def assertPandasEqual(self, df1, df2):
from odps.compat import six
from odps import types as o_types
from pandas.util.testing import assert_frame_equal
# compare column types
def get_odps_type(p_type):
for data_type, builtin_type in six.iteritems(
o_types._odps_primitive_to_builtin_types):
if issubclass(p_type.type, builtin_type):
return data_type
types1 = [get_odps_type(dt) for dt in df1.dtypes]
types2 = [get_odps_type(dt) for dt in df2.dtypes]
self.assertSequenceEqual(types1, types2)
assert_frame_equal(df1, df2, check_dtype=False)
def testJoin(self):
expr = self.odps_df.join(self.pd_df, 'name').sort('id_x')
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
def testUnion(self):
expr = self.odps_df.union(self.pd_df).sort(['id', 'name'])
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df.union(self.pd_df).sort(['id', 'name'])).values
self.assertTrue(result.equals(expected))
schema = Schema.from_lists(
[c.name for c in self.t.schema.columns if c.name != 'name'],
[c.type for c in self.t.schema.columns if c.name != 'name'],
['name'], ['string'])
t = self.odps.create_table(
'tmp_pyodps_%s' % str(uuid.uuid4()).replace('-', '_'), schema)
try:
expr = self.odps_df.union(self.pd_df)
expr.persist(t.name, create_table=False, partitions=['name'])
self.assertEqual(self.engine.execute(DataFrame(t).count()), 5)
self.engine._selecter.force_odps = False
df = DataFrame(t)
self.assertGreaterEqual(
len(
self.engine.execute(df.filter(df.name > 'a',
df.name < 'b'))), 0)
finally:
t.drop()
def testIsIn(self):
expr = self.odps_df['name'].isin(self.pd_df['name']).rename('isin')
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df['name'].isin(
self.pd_df['name']).rename('isin')).values
self.assertTrue(result.equals(expected))
expr = (self.odps_df.id + 2).isin(self.pd_df['id']).rename('isin')
res = self.engine.execute(expr)
result = self._get_result(res)
expected = [[False], [False], [True]]
self.assertEqual(result, expected)
def testMixed(self):
expr = self.odps_df.union(
self.odps_df.join(self.pd_df,
'name')[lambda x: x.name,
lambda x: x.id_x.rename('id')]).sort(
['name', 'id'])
expr = expr[expr['name'].isin(self.pd_df['name'])]
expr = expr[expr, func.rand(rtype='float').rename('rand')]
result = self.engine.execute(expr).values[['name', 'id']]
df = DataFrame(self.odps_df.to_pandas())
test_expr = df.union(
df.join(self.pd_df, 'name')[lambda x: x.name,
lambda x: x.id_x.rename('id')]).sort(
['name', 'id'])
test_expr = test_expr[test_expr['name'].isin(self.pd_df['name'])]
expected = self.pd_engine.execute(test_expr).values
self.assertTrue(result.equals(expected))
def testPandasPersist(self):
import pandas as pd, numpy as np
tmp_table_name = tn('pyodps_test_mixed_persist')
self.odps.delete_table(tmp_table_name, if_exists=True)
t = self.odps.create_table(
tmp_table_name, ('a bigint, b bigint, c bigint', 'ds string'))
t.create_partition('ds=today')
try:
pd_df = pd.DataFrame(np.arange(9).reshape(3, 3),
columns=list('abc'))
df = DataFrame(pd_df).persist(tmp_table_name,
partition='ds=today',
odps=self.odps)
self.assertPandasEqual(df[list('abc')].to_pandas(), pd_df)
finally:
self.odps.delete_table(tmp_table_name)
self.odps.to_global()
tmp_table_name = tn('pyodps_test_mixed_persist2')
self.odps.delete_table(tmp_table_name, if_exists=True)
try:
pd_df = pd.DataFrame(np.arange(9).reshape(3, 3),
columns=list('abc'))
df = DataFrame(pd_df).persist(tmp_table_name)
self.assertPandasEqual(df.to_pandas(), pd_df)
finally:
self.odps.delete_table(tmp_table_name)
def testExecuteCacheTable(self):
df = self.odps_df[self.odps_df.name == 'name1']
result = df.execute().values
self.assertEqual(len(result), 2)
self.assertTrue(context.is_cached(df))
dag = self.engine.compile(df)
calls = dag.topological_sort()
self.assertEqual(len(calls), 1)
self.assertTrue(is_source_collection(calls[0].expr))
df2 = df[:5]
result = df2.execute()
self.assertEqual(len(result), 2)
def testHandleCache(self):
df = self.pd_df['name', self.pd_df.id + 1]
df.execute()
self.assertTrue(context.is_cached(df))
df2 = df[df.id < 10]
dag = self.engine.compile(df2)
self.assertEqual(len(dag.nodes()), 1)
self.assertTrue(is_source_collection(dag.nodes()[0].expr.input))
df3 = self.pd_df[self.pd_df.id < 10].count()
i = df3.execute()
self.assertTrue(context.is_cached(df3))
df4 = df3 + 1
dag = self.engine.compile(df4)
self.assertEqual(len(dag.nodes()), 1)
self.assertIsNotNone(dag.nodes()[0].expr._fields[0].lhs.value)
self.assertEqual(df4.execute(), i + 1)
def testCacheTable(self):
self.engine._selecter.force_odps = True
df = self.odps_df.join(self.pd_df, 'name').cache()
df2 = df.sort('id_x')
dag = self.engine.compile(df2)
self.assertEqual(len(dag.nodes()), 3)
result = self.engine.execute(df2).values
df3 = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df3.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
self.assertEqual(len(self.engine._generated_table_names), 2)
table = context.get_cached(df)
self.assertEqual(len(self.engine.execute(df)), len(expected))
self.assertIs(context.get_cached(df), table)
if not isinstance(table, SeahawksTable):
self.assertEqual(context.get_cached(df).lifecycle, 1)
df4 = df[df.id_x < 3].count()
result = self.engine.execute(df4)
self.assertEqual(result, 2)
self.assertEqual(context.get_cached(df4), 2)
def testUseCache(self):
self.engine._selecter.force_odps = True
df_cache = self.odps_df[self.odps_df['name'] == 'name1'].cache()
df = df_cache[df_cache.id * 2, df_cache.exclude('id')]
self.assertEqual(len(self.engine.execute(df, head=10)), 2)
context.get_cached(df_cache).drop()
self.assertEqual(
len(self.engine.execute(df_cache['name', df_cache.id * 2],
head=10)), 2)
self.assertTrue(context.is_cached(df_cache))
self.assertTrue(
self.odps.exist_table(context.get_cached(df_cache).name))
def testHeadAndTail(self):
res = self.odps_df.head(2)
self.assertEqual(len(res), 2)
df = self.odps_df[self.odps_df['name'] == 'name1']
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertTrue(context.is_cached(df))
res = self.odps_df.tail(2)
self.assertEqual(len(res), 2)
self.assertTrue(all(it > 1 for it in res.values['id']))
self.assertEqual(len(self.odps_df.name.head(2)), 2)
self.assertEqual(len(self.odps_df.name.tail(2)), 2)
res = self.pd_df.head(1)
self.assertEqual(len(res), 1)
df = self.pd_df[self.pd_df['name'] == 'name1']
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertTrue(context.is_cached(df))
res = self.pd_df.tail(1)
self.assertEqual(len(res), 1)
self.assertEqual(res.values['id'][0], 6)
self.assertEqual(len(self.pd_df.name.head(1)), 1)
self.assertEqual(len(self.pd_df.name.tail(1)), 1)
class TunnelOnlyODPSEngine(ODPSSQLEngine):
def _do_execute(self, *args, **kwargs):
kwargs['_force_tunnel'] = True
return super(TunnelOnlyODPSEngine,
self)._do_execute(*args, **kwargs)
engine = MixedEngine(self.odps)
engine._odpssql_engine = TunnelOnlyODPSEngine(self.odps)
res = engine.execute(self.odps_df['id'], head=3)
self.assertIsNotNone(res)
self.assertEqual(sum(res.values['id']), 6)
table_name = tn('pyodps_df_mixed2')
self.odps.delete_table(table_name, if_exists=True)
table = next(self.odps_df.data_source())
table2 = self.odps.create_table(table_name, table.schema)
try:
res = DataFrame(table2).head(10)
self.assertEqual(len(res), 0)
finally:
table2.drop()
def testMapReduceWithResource(self):
pd_df2 = self.odps_df.to_pandas(wrap=True)
@output(['name', 'id'], ['string', 'int'])
def reducer(resources):
d = dict()
for r in resources[0]:
if r.name in d:
d[r.name] += r.id
else:
d[r.name] = r.id
def inner(keys):
def h(row, done):
if row.name in d:
d[row.name] += row.id
else:
d[row.name] = row.id
if done:
yield row.name, d[row.name]
return h
return inner
expr = pd_df2.map_reduce(reducer=reducer,
reducer_resources=[self.pd_df],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
odps_df2 = self.pd_df.persist(tn('pyodps_df_mixed2'), odps=self.odps)
try:
expr = self.odps_df.map_reduce(reducer=reducer,
reducer_resources=[odps_df2],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
expr = self.odps_df.map_reduce(reducer=reducer,
reducer_resources=[self.pd_df],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
expr = pd_df2.map_reduce(reducer=reducer,
reducer_resources=[odps_df2],
group='name')
result = expr.execute()
self.assertEqual(result.values['id'].sum(), 17)
finally:
next(odps_df2.data_source()).drop()
def testBloomFilter(self):
import numpy as np
data2 = [['name1'], ['name3']]
table_name = tn('pyodps_test_mixed_engine_bf_table2')
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name,
schema=Schema.from_lists(['name'],
['string']))
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, data2)
try:
expr = self.odps_df.bloom_filter('name',
expr2[:1].name,
capacity=10)
res = self.engine.execute(expr)
self.assertTrue(np.all(res['name'] != 'name2'))
finally:
table2.drop()
def testCachePersist(self):
expr = self.odps_df
data2 = [['name1', 3.2], ['name3', 2.4]]
table_name = tn('pyodps_test_mixed_engine_cp_table2')
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(
name=table_name,
schema=Schema.from_lists(['name', 'fid'], ['string', 'double']))
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, data2)
@output(expr.schema.names, expr.schema.types)
def h(row):
yield row
l = expr.filter(expr.id > 0).apply(h, axis=1).cache()
r = expr2.filter(expr2.fid > 0)
joined = l.join(r, on=['name', r.fid < 4])['id', 'fid'].cache()
output_table = tn('pyodps_test_mixed_engine_cp_output_table')
self.odps.delete_table(output_table, if_exists=True)
schema = Schema.from_lists(['id', 'fid'], ['bigint', 'double'], ['ds'],
['string'])
output_t = self.odps.create_table(output_table,
schema,
if_not_exists=True)
t = joined.persist(output_table,
partition='ds=today',
create_partition=True)
self.assertEqual(len(t.execute()), 2)
# test seahawks fallback
self.assertEqual(t.input.count().execute(), 2)
output_t.drop()
def testBigintPartitionedCache(self):
table = tn('pyodps_test_bigint_partitioned_cache')
self.odps.delete_table(table, if_exists=True)
expr = self.odps_df.persist(table, partitions=['id'])
@output(['id', 'name'], ['int', 'string'])
def handle(row):
return row.id + 1, row.name
expr = expr['tt' + expr.name, expr.id].cache()
new_expr = expr.map_reduce(mapper=handle)
res = self.engine.execute(new_expr)
self.assertEqual(len(res), 3)
def testAsync(self):
expr = self.odps_df[self.odps_df.name == 'name1']
future = self.engine.execute(expr, async_=True)
self.assertFalse(future.done())
res = future.result()
self.assertEqual(len(res), 2)
def testBatch(self):
odps_expr = self.odps_df[self.odps_df.id < 4].cache()
expr = odps_expr.join(self.pd_df, 'name').sort('id_x')
dag = self.engine.compile(expr)
self.assertEqual(len(dag.nodes()), 3)
f = self.engine.execute(expr, async_=True, n_parallel=2)
result = f.result().values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(
df.join(self.pd_df, 'name').sort('id_x')).values
self.assertTrue(result.equals(expected))
def testBatchStop(self):
self.engine._selecter.force_odps = True
expr1 = self.odps_df[self.odps_df.id < 3].cache()
expr2 = self.odps_df[self.odps_df.id > 3].cache()
expr3 = expr1.union(expr2)
self.engine.execute([expr1, expr2, expr3], n_parallel=2, async_=True)
time.sleep(2)
instance_ids = self.engine._odpssql_engine._instances
self.assertEqual(len(instance_ids), 2)
self.engine.stop()
instances = [self.odps.get_instance(i) for i in instance_ids]
[i.wait_for_completion() for i in instances]
self.assertEqual(
list(instances[0].get_task_statuses().values())[0].status,
Instance.Task.TaskStatus.CANCELLED)
self.assertEqual(
list(instances[1].get_task_statuses().values())[0].status,
Instance.Task.TaskStatus.CANCELLED)
def testFailure(self):
from odps.df.backends.errors import DagDependencyError
expr1 = self.odps_df[self.odps_df.id / 0 < 0].cache()
expr2 = expr1.count()
fs = self.engine.execute(expr2, async_=True)
self.assertRaises(DagDependencyError, fs.result)
def testAppendIDCache(self):
options.ml.dry_run = False
@output(['id1'] + self.odps_df.schema.names,
['int'] + self.odps_df.schema.types)
def h(row):
yield row
expr1 = self.odps_df.append_id(id_col='id1').apply(h, axis=1)
expr2 = self.odps_df.append_id(id_col='id2')
expr3 = expr1.join(expr2, on='id')['id1', 'id2']
self.assertEqual(len(expr3.execute()), 3)
def testAppendId(self):
options.ml.dry_run = False
expr = self.odps_df['name', ].distinct()
expr = expr.append_id(id_col='id2')
expr = expr.join(self.odps_df, on=['name'])
tablename = tn('pyodps_test_append_id_persist')
self.odps.delete_table(tablename, if_exists=True)
expr.persist(tablename, partitions=['name'], lifecycle=1)
def testHorzConcat(self):
options.ml.dry_run = False
table_name = tn('test_horz_concat_table2_xxx_yyy')
self.odps.delete_table(table_name, if_exists=True)
result_table_name = tn('test_horz_concat_result')
self.odps.delete_table(result_table_name, if_exists=True)
self.odps_df[self.odps_df.name,
(self.odps_df.id *
2).rename('ren_id')].persist(table_name)
df2 = self.odps.get_table(table_name).to_df()
df2 = df2[:3]
expr = self.odps_df.concat(df2.ren_id, axis=1)
expr.persist(result_table_name, lifecycle=1)
def testAsTypeMapReduce(self):
expr = self.odps_df[self.odps_df.exclude('id'),
self.odps_df.id.astype('float')]
expr = expr.filter(expr.id < 10)['id', 'name']
@output(['id', 'name'], ['float', 'string'])
def h(group):
def inn(row, done):
yield row
return inn
expr = expr.map_reduce(reducer=h)
expr.execute()
expr = self.odps_df[self.odps_df.exclude('id'),
self.odps_df.id.astype('float')]
expr = expr.filter(expr.id < 10).distinct('id', 'name')
@output(['id', 'name'], ['float', 'string'])
def h(group):
def inn(row, done):
yield row
return inn
expr = expr.map_reduce(reducer=h)
expr.execute()
class Test(TestBase):
def setup(self):
import pandas as pd
odps_data = [["name1", 1], ["name2", 2], ["name1", 3]]
pd_data = [["name1", 5], ["name2", 6]]
names = ["name", "id"]
types = ["string", "bigint"]
table = tn("pyodps_df_mixed")
self.odps.delete_table(table, if_exists=True)
self.t = self.odps.create_table(table, Schema.from_lists(names, types))
with self.t.open_writer() as w:
w.write([self.t.new_record(r) for r in odps_data])
self.odps_df = DataFrame(self.t)
self.pd_df = DataFrame(pd.DataFrame(pd_data, columns=names))
self.engine = MixedEngine(self.odps)
self.pd_engine = PandasEngine(self.odps)
def teardown(self):
self.t.drop()
def testGroupReduction(self):
expr = self.odps_df.select(self.odps_df, id2=self.odps_df.id.map(lambda x: x + 1))
expr = expr.groupby("name").id2.sum()
expected = [["name1", 6], ["name2", 3]]
res = self.engine.execute(expr)
result = self._get_result(res)
self.assertEqual(sorted([[r[1]] for r in expected]), sorted(result))
def assertPandasEqual(self, df1, df2):
from odps.compat import six
from odps import types as o_types
from pandas.util.testing import assert_frame_equal
# compare column types
def get_odps_type(p_type):
for data_type, builtin_type in six.iteritems(o_types._odps_primitive_to_builtin_types):
if issubclass(p_type.type, builtin_type):
return data_type
types1 = [get_odps_type(dt) for dt in df1.dtypes]
types2 = [get_odps_type(dt) for dt in df2.dtypes]
self.assertSequenceEqual(types1, types2)
assert_frame_equal(df1, df2, check_dtype=False)
def testJoin(self):
expr = self.odps_df.join(self.pd_df, "name").sort("id_x")
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df.join(self.pd_df, "name").sort("id_x")).values
self.assertTrue(result.equals(expected))
def testUnion(self):
expr = self.odps_df.union(self.pd_df).sort(["id", "name"])
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df.union(self.pd_df).sort(["id", "name"])).values
self.assertTrue(result.equals(expected))
def testIsIn(self):
expr = self.odps_df["name"].isin(self.pd_df["name"]).rename("isin")
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df["name"].isin(self.pd_df["name"]).rename("isin")).values
self.assertTrue(result.equals(expected))
def testMixed(self):
expr = self.odps_df.union(
self.odps_df.join(self.pd_df, "name")[lambda x: x.name, lambda x: x.id_x.rename("id")]
).sort(["name", "id"])
expr = expr[expr["name"].isin(self.pd_df["name"])]
result = self.engine.execute(expr).values
df = DataFrame(self.odps_df.to_pandas())
test_expr = df.union(df.join(self.pd_df, "name")[lambda x: x.name, lambda x: x.id_x.rename("id")]).sort(
["name", "id"]
)
test_expr = test_expr[test_expr["name"].isin(self.pd_df["name"])]
expected = self.pd_engine.execute(test_expr).values
self.assertTrue(result.equals(expected))
def testPandasPersist(self):
import pandas as pd, numpy as np
self.odps.to_global()
tmp_table_name = tn("pyodps_test_mixed_persist")
self.odps.delete_table(tmp_table_name, if_exists=True)
pd_df = pd.DataFrame(np.arange(9).reshape(3, 3), columns=list("abc"))
df = DataFrame(pd_df).persist(tmp_table_name)
self.assertPandasEqual(df.to_pandas(), pd_df)
self.odps.delete_table(tmp_table_name)
def testExecuteCacheTable(self):
df = self.odps_df[self.odps_df.name == "name1"]
result = df.execute().values
self.assertEqual(len(result), 2)
self.assertIsNotNone(df._cache_data)
_, new_df, cbs = self.engine._compile(df)
try:
self.assertIsNotNone(new_df._source_data)
finally:
[cb() for cb in cbs]
df2 = df[:5]
result = df2.execute()
self.assertEqual(len(result), 2)
def testHandleCache(self):
df = self.pd_df["name", self.pd_df.id + 1]
df.execute()
self.assertIsNotNone(df._cache_data)
df2 = df[df.id < 10]
_, new_df2, cbs = self.engine._compile(df2)
try:
self.assertIsNotNone(new_df2.input._source_data)
finally:
[cb() for cb in cbs]
def testCacheTable(self):
df = self.odps_df.join(self.pd_df, "name").cache()
df2 = df.sort("id_x")
dag = self.engine._compile_dag(df2)
self.assertEqual(len(dag.nodes()), 3)
result = self.engine.execute(df2).values
df3 = DataFrame(self.odps_df.to_pandas())
expected = self.pd_engine.execute(df3.join(self.pd_df, "name").sort("id_x")).values
self.assertTrue(result.equals(expected))
self.assertEqual(len(self.engine._generated_table_names), 2)
table = df._cache_data
self.assertEqual(len(df.execute()), len(expected))
self.assertIs(df._cache_data, table)
df4 = df[df.id_x < 3].count()
result = self.engine.execute(df4)
self.assertEqual(result, 2)
self.assertEqual(df4._cache_data, 2)
def testUseCache(self):
df = self.odps_df[self.odps_df["name"] == "name1"]
self.assertEqual(len(df.head(10)), 2)
df._cache_data.drop()
self.assertRaises(ODPSError, lambda: self.engine.execute(df["name", "id"]))
def plot(**_):
pass
self.assertRaises(ODPSError, lambda: df.plot(x="id", plot_func=plot))
def testPivot(self):
data = [["name1", 1, 1.0, True], ["name1", 2, 2.0, True], ["name2", 1, 3.0, False], ["name2", 3, 4.0, False]]
table_name = tn("pyodps_test_mixed_engine_pivot")
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(
name=table_name,
schema=Schema.from_lists(["name", "id", "fid", "ismale"], ["string", "bigint", "double", "boolean"]),
)
expr = DataFrame(table)
try:
self.odps.write_table(table, 0, data)
expr1 = expr.pivot(rows="id", columns="name", values="fid").distinct()
res = self.engine.execute(expr1)
result = self._get_result(res)
expected = [[1, 1.0, 3.0], [2, 2.0, None], [3, None, 4.0]]
self.assertEqual(sorted(result), sorted(expected))
expr2 = expr.pivot(rows="id", columns="name", values=["fid", "ismale"])
res = self.engine.execute(expr2)
result = self._get_result(res)
expected = [[1, 1.0, 3.0, True, False], [2, 2.0, None, True, None], [3, None, 4.0, None, False]]
self.assertEqual(sorted(result), sorted(expected))
expr3 = expr.pivot(rows="id", columns="name", values="fid")["name3"]
with self.assertRaises(ValueError) as cm:
self.engine.execute(expr3)
self.assertIn("name3", str(cm.exception))
expr4 = expr.pivot(rows="id", columns="name", values="fid")["id", "name1"]
res = self.engine.execute(expr4)
result = self._get_result(res)
expected = [[1, 1.0], [2, 2.0], [3, None]]
self.assertEqual(sorted(result), sorted(expected))
expr5 = expr.pivot(rows="id", columns="name", values="fid")
expr5 = expr5[expr5, (expr5["name1"].astype("int") + 1).rename("new_name")]
res = self.engine.execute(expr5)
result = self._get_result(res)
expected = [[1, 1.0, 3.0, 2.0], [2, 2.0, None, 3.0], [3, None, 4.0, None]]
self.assertEqual(sorted(result), sorted(expected))
expr6 = expr.pivot(rows="id", columns="name", values="fid")
expr6 = expr6.join(self.odps_df, on="id")[expr6, "name"]
res = self.engine.execute(expr6)
result = self._get_result(res)
expected = [[1, 1.0, 3.0, "name1"], [2, 2.0, None, "name2"], [3, None, 4.0, "name1"]]
self.assertEqual(sorted(result), sorted(expected))
finally:
table.drop()
def testPivotTable(self):
data = [
["name1", 1, 1.0, True],
["name1", 1, 5.0, True],
["name1", 2, 2.0, True],
["name2", 1, 3.0, False],
["name2", 3, 4.0, False],
]
table_name = tn("pyodps_test_mixed_engine_pivot_table")
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(
name=table_name,
schema=Schema.from_lists(["name", "id", "fid", "ismale"], ["string", "bigint", "double", "boolean"]),
)
expr = DataFrame(table)
try:
self.odps.write_table(table, 0, data)
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.assertEqual(sorted(result), sorted(expected))
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.assertEqual(sorted(result), sorted(expected))
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)
# TODO: fix this situation, act different compared to pandas
expected = [[2, 2], [3, 5], [None, None]]
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 = [[1, 3.0, 3.0], [2, 2.0, 0], [3, 0, 4.0]]
self.assertEqual(res.schema.names, ["id", "name1_fid_mean", "name2_fid_mean"])
self.assertEqual(result, expected)
class Agg(object):
def buffer(self):
return [0]
def __call__(self, buffer, val):
buffer[0] += val
def merge(self, buffer, pbuffer):
buffer[0] += pbuffer[0]
def getvalue(self, buffer):
return buffer[0]
aggfuncs = OrderedDict([("my_sum", Agg), ("mean", "mean")])
expr4 = expr.pivot_table(rows="id", values="fid", columns="name", fill_value=0, aggfunc=aggfuncs)
res = self.engine.execute(expr4)
result = self._get_result(res)
expected = [[1, 6.0, 3.0, 3.0, 3.0], [2, 2.0, 0, 2.0, 0], [3, 0, 4.0, 0, 4.0]]
self.assertEqual(
res.schema.names, ["id", "name1_fid_my_sum", "name2_fid_my_sum", "name1_fid_mean", "name2_fid_mean"]
)
self.assertEqual(result, expected)
finally:
table.drop()
def testExtractKV(self):
data = [
["name1", "k1=1,k2=3,k5=10", "1=5,3=7,2=1"],
["name1", "", "3=1,4=2"],
["name1", "k1=7.1,k7=8.2", "1=1,5=6"],
["name2", "k2=1.2,k3=1.5", None],
["name2", "k9=1.1,k2=1", "4=2"],
]
table_name = tn("pyodps_test_mixed_engine_extract_kv")
self.odps.delete_table(table_name, if_exists=True)
table = self.odps.create_table(
name=table_name, schema=Schema.from_lists(["name", "kv", "kv2"], ["string", "string", "string"])
)
expr = DataFrame(table)
try:
self.odps.write_table(table, 0, data)
expr1 = expr.extract_kv(columns=["kv", "kv2"], kv_delim="=")
res = self.engine.execute(expr1)
result = self._get_result(res)
expected_cols = [
"name",
"kv_k1",
"kv_k2",
"kv_k3",
"kv_k5",
"kv_k7",
"kv_k9",
"kv2_1",
"kv2_2",
"kv2_3",
"kv2_4",
"kv2_5",
]
expected = [
["name1", 1.0, 3.0, None, 10.0, None, None, 5.0, 1.0, 7.0, None, None],
["name1", None, None, None, None, None, None, None, None, 1.0, 2.0, None],
["name1", 7.1, None, None, None, 8.2, None, 1.0, None, None, None, 6.0],
["name2", None, 1.2, 1.5, None, None, None, None, None, None, None, None],
["name2", None, 1.0, None, None, None, 1.1, None, None, None, 2.0, None],
]
self.assertListEqual([c.name for c in res.columns], expected_cols)
self.assertEqual(result, expected)
finally:
table.drop()
def testHeadAndTail(self):
res = self.odps_df.head(2)
self.assertEqual(len(res), 2)
df = self.odps_df[self.odps_df["name"] == "name1"]
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertIsNotNone(df._cache_data)
res = self.odps_df.tail(2)
self.assertEqual(len(res), 2)
self.assertTrue(all(it > 1 for it in res.values["id"]))
self.assertEqual(len(self.odps_df.name.head(2)), 2)
self.assertEqual(len(self.odps_df.name.tail(2)), 2)
res = self.pd_df.head(1)
self.assertEqual(len(res), 1)
df = self.pd_df[self.pd_df["name"] == "name1"]
res = df.head(1)
self.assertEqual(len(res), 1)
self.assertIsNotNone(df._cache_data)
res = self.pd_df.tail(1)
self.assertEqual(len(res), 1)
self.assertEqual(res.values["id"][0], 6)
self.assertEqual(len(self.pd_df.name.head(1)), 1)
self.assertEqual(len(self.pd_df.name.tail(1)), 1)
class TunnelOnlyODPSEngine(ODPSEngine):
def execute(self, expr, **kw):
expr = self._pre_process(expr)
head = kw.get("head")
return self._handle_cases(expr, head=head)
engine = MixedEngine(self.odps)
engine._odpssql_engine = TunnelOnlyODPSEngine(self.odps, global_optimize=False)
res = engine.execute(self.odps_df["id"], head=3)
self.assertIsNotNone(res)
self.assertEqual(sum(res.values["id"]), 6)
table_name = tn("pyodps_df_mixed2")
self.odps.delete_table(table_name, if_exists=True)
table = next(self.odps_df.data_source())
table2 = self.odps.create_table(table_name, table.schema)
try:
res = DataFrame(table2).head(10)
self.assertEqual(len(res), 0)
finally:
table2.drop()
def testMapReduceWithResource(self):
pd_df2 = self.odps_df.to_pandas(wrap=True)
@output(["name", "id"], ["string", "int"])
def reducer(resources):
d = dict()
for r in resources[0]:
if r.name in d:
d[r.name] += r.id
else:
d[r.name] = r.id
def inner(keys):
def h(row, done):
if row.name in d:
d[row.name] += row.id
else:
d[row.name] = row.id
if done:
yield row.name, d[row.name]
return h
return inner
expr = pd_df2.map_reduce(reducer=reducer, reducer_resources=[self.pd_df], group="name")
result = expr.execute()
self.assertEqual(result.values["id"].sum(), 17)
odps_df2 = self.pd_df.persist(tn("pyodps_df_mixed2"), odps=self.odps)
try:
expr = self.odps_df.map_reduce(reducer=reducer, reducer_resources=[odps_df2], group="name")
result = expr.execute()
self.assertEqual(result.values["id"].sum(), 17)
expr = self.odps_df.map_reduce(reducer=reducer, reducer_resources=[self.pd_df], group="name")
result = expr.execute()
self.assertEqual(result.values["id"].sum(), 17)
expr = pd_df2.map_reduce(reducer=reducer, reducer_resources=[odps_df2], group="name")
result = expr.execute()
self.assertEqual(result.values["id"].sum(), 17)
finally:
next(odps_df2.data_source()).drop()
def testBloomFilter(self):
import numpy as np
data2 = [["name1"], ["name3"]]
table_name = tn("pyodps_test_mixed_engine_bf_table2")
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(name=table_name, schema=Schema.from_lists(["name"], ["string"]))
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, data2)
try:
expr = self.odps_df.bloom_filter("name", expr2[:1].name, capacity=10)
res = self.engine.execute(expr)
self.assertTrue(np.all(res["name"] != "name2"))
finally:
table2.drop()
def testCachePersist(self):
expr = self.odps_df
data2 = [["name1", 3.2], ["name3", 2.4]]
table_name = tn("pyodps_test_mixed_engine_cp_table2")
self.odps.delete_table(table_name, if_exists=True)
table2 = self.odps.create_table(
name=table_name, schema=Schema.from_lists(["name", "fid"], ["string", "double"])
)
expr2 = DataFrame(table2)
self.odps.write_table(table2, 0, data2)
@output(expr.schema.names, expr.schema.types)
def h(row):
yield row
l = expr.filter(expr.id > 0).apply(h, axis=1).cache()
r = expr2.filter(expr2.fid > 0)
joined = l.join(r, on=["name", r.fid < 4])["id", "fid"].cache()
output_table = tn("pyodps_test_mixed_engine_cp_output_table")
self.odps.delete_table(output_table, if_exists=True)
schema = Schema.from_lists(["id", "fid"], ["bigint", "double"], ["ds"], ["string"])
output_t = self.odps.create_table(output_table, schema, if_not_exists=True)
t = joined.persist(output_table, partition="ds=today", create_partition=True)
self.assertEqual(len(t.execute()), 2)
output_t.drop()
def testBigintPartitionedCache(self):
table = tn("pyodps_test_bigint_partitioned_cache")
self.odps.delete_table(table, if_exists=True)
expr = self.odps_df.persist(table, partitions=["id"])
@output(["id", "name"], ["int", "string"])
def handle(row):
return row.id + 1, row.name
expr = expr["tt" + expr.name, expr.id].cache()
new_expr = expr.map_reduce(mapper=handle)
res = self.engine.execute(new_expr)
self.assertEqual(len(res), 3)
