def test_to_hdf_lock_delays():
pytest.importorskip('tables')
df16 = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p'],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]},
index=[1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15., 16.])
a = dd.from_pandas(df16, 16)
# adding artifichial delays to make sure last tasks finish first
# that's a way to simulate last tasks finishing last
def delayed_nop(i):
if i[1] < 10:
sleep(0.1*(10-i[1]))
return i
# saving to multiple hdf nodes
with tmpfile() as fn:
a = a.apply(delayed_nop, axis=1, columns=a.columns)
a.to_hdf(fn, '/data*')
out = dd.read_hdf(fn, '/data*')
eq(df16, out)
# saving to multiple hdf files
# adding artifichial delays to make sure last tasks finish first
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
a = a.apply(delayed_nop, axis=1, columns=a.columns)
a.to_hdf(fn, '/data')
out = dd.read_hdf(fn, '/data')
eq(df16, out)
def test_to_hdf_thread():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p'],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]},
index=[1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15., 16.])
a = dd.from_pandas(df, 16)
# test single file single node
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data', get=dask.threaded.get)
out = pd.read_hdf(fn, '/data')
eq(df, out)
# test multiple files single node
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data', get=dask.threaded.get)
out = dd.read_hdf(fn, '/data')
eq(df, out)
# test single file multiple nodes
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data*', get=dask.threaded.get)
out = dd.read_hdf(fn, '/data*')
eq(df, out)
def test_to_hdf_link_optimizations():
"""testing dask link levels is correct by calculating the depth of the dask graph"""
pytest.importorskip('tables')
df16 = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p'],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]},
index=[1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15., 16.])
a = dd.from_pandas(df16, 16)
# saving to multiple hdf files, no links are needed
# expected layers: from_pandas, to_hdf, list = depth of 3
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
d = a.to_hdf(fn, '/data', compute=False)
assert dependency_depth(d.dask) == 3
# saving to a single hdf file with multiple nodes
# all subsequent nodes depend on the first
# expected layers: from_pandas, first to_hdf(creates file+node), subsequent to_hdfs, list = 4
with tmpfile() as fn:
d = a.to_hdf(fn, '/data*', compute=False)
assert dependency_depth(d.dask) == 4
# saving to a single hdf file with a single node
# every node depends on the previous node
# expected layers: from_pandas, to_hdf times npartitions(15), list = 2 + npartitions = 17
with tmpfile() as fn:
d = a.to_hdf(fn, '/data', compute=False)
assert dependency_depth(d.dask) == 2 + a.npartitions
def test_read_chunked(block):
with tmpdir() as path:
fn = os.path.join(path, '1.json')
df.to_json(fn, orient='records', lines=True)
d = dd.read_json(fn, blocksize=block, sample=10)
assert (d.npartitions > 1) or (block > 50)
assert_eq(d, df, check_index=False)
def test_hdf_globbing():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
with tmpdir() as tdir:
df.to_hdf(os.path.join(tdir, 'one.h5'), '/foo/data', format='table')
df.to_hdf(os.path.join(tdir, 'two.h5'), '/bar/data', format='table')
df.to_hdf(os.path.join(tdir, 'two.h5'), '/foo/data', format='table')
with dask.set_options(get=dask.get):
res = dd.read_hdf(os.path.join(tdir, 'one.h5'), '/*/data',
chunksize=2)
assert res.npartitions == 2
tm.assert_frame_equal(res.compute(), df)
res = dd.read_hdf(os.path.join(tdir, 'one.h5'), '/*/data',
chunksize=2, start=1, stop=3)
expected = pd.read_hdf(os.path.join(tdir, 'one.h5'), '/foo/data',
start=1, stop=3)
tm.assert_frame_equal(res.compute(), expected)
res = dd.read_hdf(os.path.join(tdir, 'two.h5'), '/*/data', chunksize=2)
assert res.npartitions == 2 + 2
tm.assert_frame_equal(res.compute(), pd.concat([df] * 2))
res = dd.read_hdf(os.path.join(tdir, '*.h5'), '/foo/data', chunksize=2)
assert res.npartitions == 2 + 2
tm.assert_frame_equal(res.compute(), pd.concat([df] * 2))
res = dd.read_hdf(os.path.join(tdir, '*.h5'), '/*/data', chunksize=2)
assert res.npartitions == 2 + 2 + 2
tm.assert_frame_equal(res.compute(), pd.concat([df] * 3))
def test_to_textfiles_name_function_preserves_order():
seq = ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p']
b = db.from_sequence(seq, npartitions=16)
with tmpdir() as dn:
b.to_textfiles(dn)
out = db.read_text(os.path.join(dn, "*"), encoding='ascii').map(str).map(str.strip).compute()
assert seq == out
def random_images(n, shape):
with tmpdir() as dirname:
for i in range(n):
fn = os.path.join(dirname, 'image.%d.png' % i)
x = np.random.randint(0, 255, size=shape).astype('i1')
imsave(fn, x)
yield os.path.join(dirname, '*.png')
def test_to_hdf_multiple_files():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
a = dd.from_pandas(df, 2)
df16 = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i',
'j', 'k', 'l', 'm', 'n', 'o', 'p'],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14,
15, 16]},
index=[1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11.,
12., 13., 14., 15., 16.])
b = dd.from_pandas(df16, 16)
# saving to multiple files
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data')
out = dd.read_hdf(fn, '/data')
assert_eq(df, out)
# saving to multiple files making sure order is kept
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.h5')
b.to_hdf(fn, '/data')
out = dd.read_hdf(fn, '/data')
assert_eq(df16, out)
# saving to multiple files with custom name_function
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data', name_function=lambda i: 'a' * (i + 1))
out = dd.read_hdf(fn, '/data')
assert_eq(df, out)
out = pd.read_hdf(os.path.join(dn, 'data_a.h5'), '/data')
tm.assert_frame_equal(out, df.iloc[:2])
out = pd.read_hdf(os.path.join(dn, 'data_aa.h5'), '/data')
tm.assert_frame_equal(out, df.iloc[2:])
# test hdf object
with tmpfile('h5') as fn:
with pd.HDFStore(fn) as hdf:
a.to_hdf(hdf, '/data*')
out = dd.read_hdf(fn, '/data*')
assert_eq(df, out)
def test_write_json_basic(orient):
with tmpdir() as path:
fn = os.path.join(path, '1.json')
df.to_json(fn, orient=orient, lines=False)
actual = dd.read_json(fn, orient=orient, lines=False)
out = actual.compute()
if orient == 'values':
out.columns = list(df.columns)
assert_eq(out, df)
def test_to_csv_multiple_files_cornercases():
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]})
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
with pytest.raises(ValueError):
fn = os.path.join(dn, "data_*_*.csv")
a.to_csv(fn)
df16 = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p'],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]})
a = dd.from_pandas(df16, 16)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.csv')
a.to_csv(fn, index=False)
result = dd.read_csv(fn).compute().reset_index(drop=True)
eq(result, df16)
# test handling existing files when links are optimized out
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_1.csv')
a.to_csv(fn, index=False)
fn = os.path.join(dn, 'data_*.csv')
a.to_csv(fn, mode='w', index=False)
result = dd.read_csv(fn).compute().reset_index(drop=True)
eq(result, df)
# test handling existing files when links are optimized out
a = dd.from_pandas(df16, 16)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_01.csv')
a.to_csv(fn, index=False)
fn = os.path.join(dn, 'data_*.csv')
a.to_csv(fn, mode='w', index=False)
result = dd.read_csv(fn).compute().reset_index(drop=True)
eq(result, df16)
# test handling existing files when mode isn't 'w'
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.csv')
with pytest.raises(ValueError):
a.to_csv(fn, mode='a')
def test_to_csv_simple():
df0 = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
df = dd.from_pandas(df0, npartitions=2)
with tmpdir() as dir:
dir = str(dir)
df.to_csv(dir)
assert os.listdir(dir)
result = dd.read_csv(os.path.join(dir, '*')).compute()
assert (result.x.values == df0.x.values).all()
def test_roundtrip(df, write_kwargs, read_kwargs):
with tmpdir() as tmp:
tmp = str(tmp)
if df.index.name is None:
df.index.name = 'index'
ddf = dd.from_pandas(df, npartitions=2)
to_parquet(tmp, ddf, **write_kwargs)
ddf2 = read_parquet(tmp, index=df.index.name, **read_kwargs)
assert_eq(ddf, ddf2)
def test_to_csv_series():
df0 = pd.Series(['a', 'b', 'c', 'd'], index=[1., 2., 3., 4.])
df = dd.from_pandas(df0, npartitions=2)
with tmpdir() as dir:
dir = str(dir)
df.to_csv(dir, header=False)
assert os.listdir(dir)
result = dd.read_csv(os.path.join(dir, '*'), header=None,
names=['x']).compute()
assert (result.x == df0).all()
def test_json_compressed(compression):
if compression == 'xz' and lzma is None:
pytest.skip(
"LZMA not available. Please install backports.lzma on Python 2."
)
with tmpdir() as path:
dd.to_json(ddf, path, compression=compression)
actual = dd.read_json(os.path.join(path, '*'),
compression=compression)
assert_eq(df, actual.compute(), check_index=False)
def test_zarr_distributed_roundtrip(loop):
da = pytest.importorskip('dask.array')
zarr = pytest.importorskip('zarr')
assert_eq = da.utils.assert_eq
with cluster() as (s, [a, b]):
with Client(s['address'], loop=loop) as c:
with tmpdir() as d:
a = da.zeros((3, 3), chunks=(1, 1))
a.to_zarr(d)
a2 = da.from_zarr(d)
assert_eq(a, a2)
assert a2.chunks == a.chunks
def test_to_hdf_modes_multiple_files():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
# appending a single partition to existing data
a = dd.from_pandas(df, 1)
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
a.to_hdf(os.path.join(dn, 'data2'), '/data')
a.to_hdf(fn, '/data', mode='a')
out = dd.read_hdf(fn, '/data*')
eq(df.append(df), out)
# appending two partitions to existing data
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
a.to_hdf(os.path.join(dn, 'data2'), '/data')
a.to_hdf(fn, '/data', mode='a')
out = dd.read_hdf(fn, '/data')
eq(df.append(df), out)
# overwriting a file with two partitions
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
a.to_hdf(os.path.join(dn, 'data1'), '/data')
a.to_hdf(fn, '/data', mode='w')
out = dd.read_hdf(fn, '/data')
eq(df, out)
# overwriting a single partition, keeping other partitions
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
a.to_hdf(os.path.join(dn, 'data1'), '/data')
a.to_hdf(fn, '/data', mode='a', append=False)
out = dd.read_hdf(fn, '/data')
eq(df.append(df), out)
def test_hdf_file_list():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
with tmpdir() as tdir:
df.iloc[:2].to_hdf(os.path.join(tdir, 'one.h5'), 'dataframe', format='table')
df.iloc[2:].to_hdf(os.path.join(tdir, 'two.h5'), 'dataframe', format='table')
with dask.set_options(get=dask.get):
input_files = [os.path.join(tdir, 'one.h5'), os.path.join(tdir, 'two.h5')]
res = dd.read_hdf(input_files, 'dataframe')
tm.assert_frame_equal(res.compute(), df)
def test_categorical():
with tmpdir() as tmp:
df = pd.DataFrame({'x': ['a', 'b', 'c'] * 100},
dtype='category')
ddf = dd.from_pandas(df, npartitions=3)
to_parquet(tmp, ddf)
ddf2 = read_parquet(tmp, categories=['x'])
assert ddf2.x.cat.categories.tolist() == ['a', 'b', 'c']
ddf2.loc[:1000].compute()
df.index.name = 'index' # defaults to 'index' in this case
assert assert_eq(df, ddf2)
def test_to_csv_header_empty_dataframe(header, expected):
dfe = pd.DataFrame({'x': [],
'y': []})
ddfe = dd.from_pandas(dfe, npartitions=1)
with tmpdir() as dn:
ddfe.to_csv(os.path.join(dn, "fooe*.csv"), index=False, header=header)
assert not os.path.exists(os.path.join(dn, "fooe1.csv"))
filename = os.path.join(dn, 'fooe0.csv')
with open(filename, 'r') as fp:
line = fp.readline()
assert line == expected
os.remove(filename)
def test_visualize():
pytest.importorskip('graphviz')
with tmpdir() as d:
x = da.arange(5, chunks=2)
x.visualize(filename=os.path.join(d, 'mydask'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
x.visualize(filename=os.path.join(d, 'mydask.pdf'))
assert os.path.exists(os.path.join(d, 'mydask.pdf'))
visualize(x, 1, 2, filename=os.path.join(d, 'mydask.png'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
dsk = {'a': 1, 'b': (add, 'a', 2), 'c': (mul, 'a', 1)}
visualize(x, dsk, filename=os.path.join(d, 'mydask.png'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
def test_to_csv():
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]})
for npartitions in [1, 2]:
a = dd.from_pandas(df, npartitions)
with tmpdir() as dn:
a.to_csv(dn, index=False)
result = dd.read_csv(os.path.join(dn, '*')).compute().reset_index(drop=True)
assert_eq(result, df)
with tmpdir() as dn:
r = a.to_csv(dn, index=False, compute=False)
dask.compute(*r, scheduler='sync')
result = dd.read_csv(os.path.join(dn, '*')).compute().reset_index(drop=True)
assert_eq(result, df)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.csv')
a.to_csv(fn, index=False)
result = dd.read_csv(fn).compute().reset_index(drop=True)
assert_eq(result, df)
def test_to_textfiles_inputs():
B = db.from_sequence(['abc', '123', 'xyz'], npartitions=2)
with tmpfile() as a:
with tmpfile() as b:
B.to_textfiles([a, b])
assert os.path.exists(a)
assert os.path.exists(b)
with tmpdir() as dirname:
B.to_textfiles(dirname)
assert os.path.exists(dirname)
assert os.path.exists(os.path.join(dirname, '0.part'))
assert raises(ValueError, lambda: B.to_textfiles(5))
def test_to_textfiles(ext, myopen):
b = db.from_sequence(['abc', '123', 'xyz'], npartitions=2)
with tmpdir() as dir:
c = b.to_textfiles(os.path.join(dir, '*.' + ext), compute=False)
dask.compute(*c, get=dask.get)
assert os.path.exists(os.path.join(dir, '1.' + ext))
f = myopen(os.path.join(dir, '1.' + ext), 'rb')
text = f.read()
if hasattr(text, 'decode'):
text = text.decode()
assert 'xyz' in text
f.close()
def test_compression_multiple_files():
with tmpdir() as tdir:
f = gzip.open(os.path.join(tdir, 'a.csv.gz'), 'wb')
f.write(text.encode())
f.close()
f = gzip.open(os.path.join(tdir, 'b.csv.gz'), 'wb')
f.write(text.encode())
f.close()
df = dd.read_csv(os.path.join(tdir, '*.csv.gz'), compression='gzip')
assert len(df.compute()) == (len(text.split('\n')) - 1) * 2
def test_append_wo_index():
"""Test append with write_index=False."""
with tmpdir() as tmp:
df = pd.DataFrame({'i32': np.arange(1000, dtype=np.int32),
'i64': np.arange(1000, dtype=np.int64),
'f': np.arange(1000, dtype=np.float64),
'bhello': np.random.choice(['hello', 'you', 'people'],
size=1000).astype("O")})
half = len(df) // 2
ddf1 = dd.from_pandas(df.iloc[:half], chunksize=100)
ddf2 = dd.from_pandas(df.iloc[half:], chunksize=100)
ddf1.to_parquet(tmp)
with pytest.raises(ValueError) as excinfo:
ddf2.to_parquet(tmp, write_index=False, append=True)
assert 'Appended columns' in str(excinfo.value)
with tmpdir() as tmp:
ddf1.to_parquet(tmp, write_index=False)
ddf2.to_parquet(tmp, write_index=False, append=True)
ddf3 = read_parquet(tmp, index='f')
assert_eq(df.set_index('f'), ddf3)
def test_to_textfiles_encoding():
b = db.from_sequence([u'汽车', u'苹果', u'天气'], npartitions=2)
for ext, myopen in [('gz', GzipFile), ('bz2', BZ2File), ('', open)]:
with tmpdir() as dir:
c = b.to_textfiles(os.path.join(dir, '*.' + ext), encoding='gb18030', compute=False)
c.compute(get=dask.get)
assert os.path.exists(os.path.join(dir, '1.' + ext))
f = myopen(os.path.join(dir, '1.' + ext), 'rb')
text = f.read()
if hasattr(text, 'decode'):
text = text.decode('gb18030')
assert u'天气' in text
f.close()
def test_to_textfiles():
b = db.from_sequence(['abc', '123', 'xyz'], npartitions=2)
for ext, myopen in [('gz', GzipFile), ('bz2', BZ2File), ('', open)]:
with tmpdir() as dir:
c = b.to_textfiles(os.path.join(dir, '*.' + ext), compute=False)
c.compute(get=dask.get)
assert os.path.exists(os.path.join(dir, '1.' + ext))
f = myopen(os.path.join(dir, '1.' + ext), 'rb')
text = f.read()
if hasattr(text, 'decode'):
text = text.decode()
assert 'xyz' in text
f.close()
def test_ordering():
with tmpdir() as tmp:
tmp = str(tmp)
df = pd.DataFrame({'a': [1, 2, 3],
'b': [10, 20, 30],
'c': [100, 200, 300]},
index=pd.Index([-1, -2, -3], name='myindex'),
columns=['c', 'a', 'b'])
ddf = dd.from_pandas(df, npartitions=2)
to_parquet(tmp, ddf)
pf = fastparquet.ParquetFile(tmp)
assert pf.columns == ['myindex', 'c', 'a', 'b']
ddf2 = read_parquet(tmp, index='myindex')
assert_eq(ddf, ddf2)
def test_to_textfiles_encoding():
b = db.from_sequence([u'汽车', u'苹果', u'天气'], npartitions=2)
for ext, myopen in [('gz', GzipFile), ('bz2', BZ2File), ('', open)]:
with tmpdir() as dir:
c = b.to_textfiles(os.path.join(dir, '*.' + ext),
encoding='gb18030',
compute=False)
assert c.npartitions == b.npartitions
c.compute(get=dask.get)
assert os.path.exists(os.path.join(dir, '1.' + ext))
f = myopen(os.path.join(dir, '1.' + ext), 'rb')
text = f.read()
if hasattr(text, 'decode'):
text = text.decode('gb18030')
assert u'天气' in text
f.close()
def test_to_csv_with_get():
from dask.multiprocessing import get as mp_get
flag = [False]
def my_get(*args, **kwargs):
flag[0] = True
return mp_get(*args, **kwargs)
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]})
ddf = dd.from_pandas(df, npartitions=2)
with tmpdir() as dn:
ddf.to_csv(dn, index=False, scheduler=my_get)
assert flag[0]
result = dd.read_csv(os.path.join(dn, '*')).compute().reset_index(drop=True)
assert_eq(result, df)
def test_to_csv_line_ending():
df = pd.DataFrame({"x": [0]})
ddf = dd.from_pandas(df, npartitions=1)
expected = {b"0\r\n", b"0\n"} # either/or
# For comparison...
# unexpected = {b'0\r\r\n'}
# This test addresses GH4809, and checks that only (at most) one
# '\r' character is written per line when writing to csv.
# In case it's correct (on UNIX) to have no '\r' at all, this test
# considers either '\r\n' or '\n' as appropriate line endings,
# but not '\r\r\n'.
with tmpdir() as dn:
ddf.to_csv(os.path.join(dn, "foo*.csv"), header=False, index=False)
filename = os.path.join(dn, "foo0.csv")
with open(filename, "rb") as f:
raw = f.read()
assert raw in expected
def test_to_json_with_get():
from dask.multiprocessing import get as mp_get
flag = [False]
def my_get(*args, **kwargs):
flag[0] = True
return mp_get(*args, **kwargs)
df = pd.DataFrame({"x": ["a", "b", "c", "d"], "y": [1, 2, 3, 4]})
ddf = dd.from_pandas(df, npartitions=2)
with tmpdir() as dn:
ddf.to_json(dn, compute_kwargs={"scheduler": my_get})
assert flag[0]
result = dd.read_json(os.path.join(dn, "*"))
assert_eq(result, df, check_index=False)
def test_to_hdf_exceptions():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
a = dd.from_pandas(df, 1)
# triggering too many asterisks error
with tmpdir() as dn:
with pytest.raises(ValueError):
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data_*')
# triggering too many asterisks error
with tmpfile() as fn:
with pd.HDFStore(fn) as hdf:
with pytest.raises(ValueError):
a.to_hdf(hdf, '/data_*_*')
def test_to_fmt_warns():
pytest.importorskip('tables')
df16 = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p'],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]},
index=[1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15., 16.])
a = dd.from_pandas(df16, 16)
# testing warning when breaking order
with tmpfile('h5') as fn:
with pytest.warns(None):
a.to_hdf(fn, '/data*', name_function=str)
# testing warning when breaking order
with tmpdir() as dn:
with pytest.warns(None):
fn = os.path.join(dn, "data_*.csv")
a.to_csv(fn, name_function=str)
def test_to_hdf_exceptions():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
a = dd.from_pandas(df, 1)
# triggering too many asterisks error
with tmpdir() as dn:
with pytest.raises(ValueError):
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data_*')
# triggering too many asterisks error
with tmpfile() as fn:
with pd.HDFStore(fn) as hdf:
with pytest.raises(ValueError):
a.to_hdf(hdf, '/data_*_*')
def test_to_csv_with_get():
from dask.multiprocessing import get as mp_get
flag = [False]
def my_get(*args, **kwargs):
flag[0] = True
return mp_get(*args, **kwargs)
df = pd.DataFrame({"x": ["a", "b", "c", "d"], "y": [1, 2, 3, 4]})
ddf = dd.from_pandas(df, npartitions=2)
with tmpdir() as dn:
ddf.to_csv(dn, index=False, scheduler=my_get)
assert flag[0]
result = dd.read_csv(os.path.join(dn, "*")).compute().reset_index(drop=True)
assert_eq(result, df)
def test_append():
"""Test that appended parquet equal to the original one."""
with tmpdir() as tmp:
df = pd.DataFrame({'i32': np.arange(1000, dtype=np.int32),
'i64': np.arange(1000, dtype=np.int64),
'f': np.arange(1000, dtype=np.float64),
'bhello': np.random.choice(['hello', 'yo', 'people'],
size=1000).astype("O")})
df.index.name = 'index'
half = len(df) // 2
ddf1 = dd.from_pandas(df.iloc[:half], chunksize=100)
ddf2 = dd.from_pandas(df.iloc[half:], chunksize=100)
ddf1.to_parquet(tmp)
ddf2.to_parquet(tmp, append=True)
ddf3 = read_parquet(tmp)
assert_eq(df, ddf3)
def test_visualize():
pytest.importorskip("graphviz")
X, y = make_classification(n_samples=100, n_classes=2, flip_y=0.2, random_state=0)
clf = SVC(random_state=0, gamma="auto")
grid = {"C": [0.1, 0.5, 0.9]}
gs = dcv.GridSearchCV(clf, param_grid=grid).fit(X, y)
assert hasattr(gs, "dask_graph_")
with tmpdir() as d:
gs.visualize(filename=os.path.join(d, "mydask"))
assert os.path.exists(os.path.join(d, "mydask.png"))
# Doesn't work if not fitted
gs = dcv.GridSearchCV(clf, param_grid=grid)
with pytest.raises(NotFittedError):
gs.visualize()
def test_to_hdf_exceptions():
pytest.importorskip("tables")
df = pd.DataFrame(
{"x": ["a", "b", "c", "d"], "y": [1, 2, 3, 4]}, index=[1.0, 2.0, 3.0, 4.0]
)
a = dd.from_pandas(df, 1)
# triggering too many asterisks error
with tmpdir() as dn:
with pytest.raises(ValueError):
fn = os.path.join(dn, "data_*.h5")
a.to_hdf(fn, "/data_*")
# triggering too many asterisks error
with tmpfile() as fn:
with pd.HDFStore(fn) as hdf:
with pytest.raises(ValueError):
a.to_hdf(hdf, "/data_*_*")
def test_to_textfiles_encoding():
b = db.from_sequence([u"汽车", u"苹果", u"天气"], npartitions=2)
for ext, myopen in [("gz", GzipFile), ("bz2", BZ2File), ("", open)]:
if ext == "bz2" and PY2:
continue
with tmpdir() as dir:
c = b.to_textfiles(
os.path.join(dir, "*." + ext), encoding="gb18030", compute=False
)
dask.compute(*c)
assert os.path.exists(os.path.join(dir, "1." + ext))
f = myopen(os.path.join(dir, "1." + ext), "rb")
text = f.read()
if hasattr(text, "decode"):
text = text.decode("gb18030")
assert u"天气" in text
f.close()
def test_visualize():
pytest.importorskip('graphviz')
X, y = make_classification(n_samples=100, n_classes=2, flip_y=.2,
random_state=0)
clf = SVC(random_state=0)
grid = {'C': [.1, .5, .9]}
gs = dcv.GridSearchCV(clf, grid).fit(X, y)
assert hasattr(gs, 'dask_graph_')
with tmpdir() as d:
gs.visualize(filename=os.path.join(d, 'mydask'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
# Doesn't work if not fitted
gs = dcv.GridSearchCV(clf, grid)
with pytest.raises(NotFittedError):
gs.visualize()
def test_ordering():
with tmpdir() as tmp:
tmp = str(tmp)
df = pd.DataFrame(
{
'a': [1, 2, 3],
'b': [10, 20, 30],
'c': [100, 200, 300]
},
index=pd.Index([-1, -2, -3], name='myindex'),
columns=['c', 'a', 'b'])
ddf = dd.from_pandas(df, npartitions=2)
to_parquet(tmp, ddf)
pf = fastparquet.ParquetFile(tmp)
assert pf.columns == ['myindex', 'c', 'a', 'b']
ddf2 = read_parquet(tmp, index='myindex')
assert_eq(ddf, ddf2)
def test_append_with_partition():
with tmpdir() as tmp:
df0 = pd.DataFrame({'lat': np.arange(0, 10), 'lon': np.arange(10, 20),
'value': np.arange(100, 110)})
df0.index.name = 'index'
df1 = pd.DataFrame({'lat': np.arange(10, 20), 'lon': np.arange(10, 20),
'value': np.arange(120, 130)})
df1.index.name = 'index'
dd_df0 = dd.from_pandas(df0, npartitions=1)
dd_df1 = dd.from_pandas(df1, npartitions=1)
dd.to_parquet(tmp, dd_df0, partition_on=['lon'])
dd.to_parquet(tmp, dd_df1, partition_on=['lon'], append=True,
ignore_divisions=True)
out = dd.read_parquet(tmp).compute()
out['lon'] = out.lon.astype('int64') # just to pass assert
# sort required since partitioning breaks index order
assert_eq(out.sort_values('value'), pd.concat([df0, df1])[out.columns],
check_index=False)
def test_visualize():
pytest.importorskip('graphviz')
with tmpdir() as d:
x = da.arange(5, chunks=2)
x.visualize(filename=os.path.join(d, 'mydask'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
x.visualize(filename=os.path.join(d, 'mydask.pdf'))
assert os.path.exists(os.path.join(d, 'mydask.pdf'))
visualize(x, 1, 2, filename=os.path.join(d, 'mydask.png'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
dsk = {'a': 1, 'b': (add, 'a', 2), 'c': (mul, 'a', 1)}
visualize(x, dsk, filename=os.path.join(d, 'mydask.png'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
x = Tuple(dsk, ['a', 'b', 'c'])
visualize(x, filename=os.path.join(d, 'mydask.png'))
assert os.path.exists(os.path.join(d, 'mydask.png'))
def test_visualize():
pytest.importorskip("graphviz")
with tmpdir() as d:
x = da.arange(5, chunks=2)
x.visualize(filename=os.path.join(d, "mydask"))
assert os.path.exists(os.path.join(d, "mydask.png"))
x.visualize(filename=os.path.join(d, "mydask.pdf"))
assert os.path.exists(os.path.join(d, "mydask.pdf"))
visualize(x, 1, 2, filename=os.path.join(d, "mydask.png"))
assert os.path.exists(os.path.join(d, "mydask.png"))
dsk = {"a": 1, "b": (add, "a", 2), "c": (mul, "a", 1)}
visualize(x, dsk, filename=os.path.join(d, "mydask.png"))
assert os.path.exists(os.path.join(d, "mydask.png"))
x = Tuple(dsk, ["a", "b", "c"])
visualize(x, filename=os.path.join(d, "mydask.png"))
assert os.path.exists(os.path.join(d, "mydask.png"))
def test_append_different_columns():
"""Test raising of error when non equal columns."""
with tmpdir() as tmp:
df1 = pd.DataFrame({'i32': np.arange(100, dtype=np.int32)})
df2 = pd.DataFrame({'i64': np.arange(100, dtype=np.int64)})
df3 = pd.DataFrame({'i32': np.arange(100, dtype=np.int64)})
ddf1 = dd.from_pandas(df1, chunksize=2)
ddf2 = dd.from_pandas(df2, chunksize=2)
ddf3 = dd.from_pandas(df3, chunksize=2)
ddf1.to_parquet(tmp)
with pytest.raises(ValueError) as excinfo:
ddf2.to_parquet(tmp, append=True)
assert 'Appended columns' in str(excinfo.value)
with pytest.raises(ValueError) as excinfo:
ddf3.to_parquet(tmp, append=True)
assert 'Appended dtypes' in str(excinfo.value)
def dir_server():
with tmpdir() as d:
for fn in files:
with open(os.path.join(d, fn), "wb") as f:
f.write(b"a" * 10000)
cmd = [sys.executable, "-m", "http.server", "8999"]
p = subprocess.Popen(cmd, cwd=d)
timeout = 10
while True:
try:
requests.get("http://localhost:8999")
break
except requests.exceptions.ConnectionError as e:
time.sleep(0.1)
timeout -= 0.1
if timeout < 0:
raise RuntimeError("Server did not appear") from e
yield d
p.terminate()
def test_append_overlapping_divisions():
"""Test raising of error when divisions overlapping."""
with tmpdir() as tmp:
df = pd.DataFrame({'i32': np.arange(1000, dtype=np.int32),
'i64': np.arange(1000, dtype=np.int64),
'f': np.arange(1000, dtype=np.float64),
'bhello': np.random.choice(
['hello', 'yo', 'people'],
size=1000).astype("O")})
half = len(df) // 2
ddf1 = dd.from_pandas(df.iloc[:half], chunksize=100)
ddf2 = dd.from_pandas(df.iloc[half - 10:], chunksize=100)
ddf1.to_parquet(tmp)
with pytest.raises(ValueError) as excinfo:
ddf2.to_parquet(tmp, append=True)
assert 'Appended divisions' in str(excinfo.value)
ddf2.to_parquet(tmp, append=True, ignore_divisions=True)
def test_visualize():
pytest.importorskip("graphviz")
pytest.importorskip("ipycytoscape")
with tmpdir() as d:
x = da.arange(5, chunks=2)
x.visualize(filename=os.path.join(d, "mydask"))
assert os.path.exists(os.path.join(d, "mydask.png"))
x.visualize(filename=os.path.join(d, "mydask.pdf"))
assert os.path.exists(os.path.join(d, "mydask.pdf"))
visualize(x, 1, 2, filename=os.path.join(d, "mydask.png"))
assert os.path.exists(os.path.join(d, "mydask.png"))
dsk = {"a": 1, "b": (add, "a", 2), "c": (mul, "a", 1)}
visualize(x, dsk, filename=os.path.join(d, "mydask.png"))
assert os.path.exists(os.path.join(d, "mydask.png"))
x = Tuple(dsk, ["a", "b", "c"])
visualize(x, filename=os.path.join(d, "mydask.png"))
assert os.path.exists(os.path.join(d, "mydask.png"))
x = Tuple(dsk, ["a", "b", "c"])
visualize(x, filename=os.path.join(d, "cyt"), engine="cytoscape")
assert os.path.exists(os.path.join(d, "cyt.html"))
visualize(x,
filename=os.path.join(d, "cyt2.html"),
engine="ipycytoscape")
assert os.path.exists(os.path.join(d, "cyt2.html"))
with dask.config.set(visualization__engine="cytoscape"):
visualize(x, filename=os.path.join(d, "cyt3.html"))
assert os.path.exists(os.path.join(d, "cyt3.html"))
with pytest.raises(ValueError, match="not-real"):
visualize(x, engine="not-real")
# To see if visualize() works when the filename parameter is set to None
# If the function raises an error, the test will fail
x.visualize(filename=None)
def test_to_csv():
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'], 'y': [1, 2, 3, 4]})
for npartitions in [1, 2]:
a = dd.from_pandas(df, npartitions)
with tmpfile('csv') as fn:
a.to_csv(fn, index=False)
result = dd.read_csv(fn).compute().reset_index(drop=True)
eq(result, df)
with tmpfile('csv') as fn:
r = a.to_csv(fn, index=False, compute=False)
r.compute()
result = dd.read_csv(fn).compute().reset_index(drop=True)
eq(result, df)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.csv')
a.to_csv(fn, index=False)
result = dd.read_csv(fn).compute().reset_index(drop=True)
eq(result, df)
def test_reading_empty_csv_files_with_path():
with tmpdir() as tdir:
for k, content in enumerate(["0, 1, 2", "", "6, 7, 8"]):
with open(os.path.join(tdir, str(k) + ".csv"), "w") as file:
file.write(content)
result = dd.read_csv(
os.path.join(tdir, "*.csv"),
include_path_column=True,
converters={"path": parse_filename},
names=["A", "B", "C"],
).compute()
df = pd.DataFrame(
{
"A": [0, 6],
"B": [1, 7],
"C": [2, 8],
"path": ["0.csv", "2.csv"],
}
)
df["path"] = df["path"].astype("category")
assert_eq(result, df, check_index=False)
def test_to_csv_header(header, header_first_partition_only, expected_first, expected_next):
partition_count = 2
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd', 'e', 'f'],
'y': [1, 2, 3, 4, 5, 6]})
ddf = dd.from_pandas(df, npartitions=partition_count)
with tmpdir() as dn:
# Test NO header case
# (header=False, header_first_chunk_only not passed)
ddf.to_csv(os.path.join(dn, "fooa*.csv"), index=False, header=header,
header_first_partition_only=header_first_partition_only)
filename = os.path.join(dn, 'fooa0.csv')
with open(filename, 'r') as fp:
line = fp.readline()
assert line == expected_first
os.remove(filename)
filename = os.path.join(dn, 'fooa1.csv')
with open(filename, 'r') as fp:
line = fp.readline()
assert line == expected_next
os.remove(filename)
def test_categorical():
with tmpdir() as tmp:
df = pd.DataFrame({'x': ['a', 'b', 'c'] * 100}, dtype='category')
ddf = dd.from_pandas(df, npartitions=3)
to_parquet(tmp, ddf)
ddf2 = read_parquet(tmp, categories=['x'])
assert ddf2.compute().x.cat.categories.tolist() == ['a', 'b', 'c']
# autocat
ddf2 = read_parquet(tmp)
assert ddf2.compute().x.cat.categories.tolist() == ['a', 'b', 'c']
ddf2.loc[:1000].compute()
df.index.name = 'index' # defaults to 'index' in this case
assert assert_eq(df, ddf2)
# dereference cats
ddf2 = read_parquet(tmp, categories=[])
ddf2.loc[:1000].compute()
assert (df.x == ddf2.x).all()
def test_to_hdf_lock_delays():
pytest.importorskip('tables')
df16 = pd.DataFrame(
{
'x': [
'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l',
'm', 'n', 'o', 'p'
],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]
},
index=[
1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15.,
16.
])
a = dd.from_pandas(df16, 16)
# adding artifichial delays to make sure last tasks finish first
# that's a way to simulate last tasks finishing last
def delayed_nop(i):
if i[1] < 10:
sleep(0.1 * (10 - i[1]))
return i
# saving to multiple hdf nodes
with tmpfile() as fn:
a = a.apply(delayed_nop, axis=1, columns=a.columns)
a.to_hdf(fn, '/data*')
out = dd.read_hdf(fn, '/data*')
assert_eq(df16, out)
# saving to multiple hdf files
# adding artifichial delays to make sure last tasks finish first
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
a = a.apply(delayed_nop, axis=1, columns=a.columns)
a.to_hdf(fn, '/data')
out = dd.read_hdf(fn, '/data')
assert_eq(df16, out)
def dir_server():
with tmpdir() as d:
for fn in files:
with open(os.path.join(d, fn), 'wb') as f:
f.write(b'a' * 10000)
if PY2:
cmd = [sys.executable, '-m', 'SimpleHTTPServer', '8999']
else:
cmd = [sys.executable, '-m', 'http.server', '8999']
p = subprocess.Popen(cmd, cwd=d)
timeout = 10
while True:
try:
requests.get('http://localhost:8999')
break
except requests.exceptions.ConnectionError:
time.sleep(0.1)
timeout -= 0.1
if timeout < 0:
raise RuntimeError('Server did not appear')
yield d
p.terminate()
def test_local():
with tmpdir() as tmp:
tmp = str(tmp)
data = pd.DataFrame({'i32': np.arange(1000, dtype=np.int32),
'i64': np.arange(1000, dtype=np.int64),
'f': np.arange(1000, dtype=np.float64),
'bhello': np.random.choice(['hello', 'you', 'people'], size=1000).astype("O")})
df = dd.from_pandas(data, chunksize=500)
to_parquet(tmp, df, write_index=False, object_encoding='utf8')
files = os.listdir(tmp)
assert '_metadata' in files
assert 'part.0.parquet' in files
df2 = read_parquet(tmp, index=False)
assert len(df2.divisions) > 1
out = df2.compute(get=dask.get).reset_index()
for column in df.columns:
assert (data[column] == out[column]).all()
def test_hdf_file_list():
pytest.importorskip("tables")
df = pd.DataFrame({
"x": ["a", "b", "c", "d"],
"y": [1, 2, 3, 4]
},
index=[1.0, 2.0, 3.0, 4.0])
with tmpdir() as tdir:
df.iloc[:2].to_hdf(os.path.join(tdir, "one.h5"),
"dataframe",
format="table")
df.iloc[2:].to_hdf(os.path.join(tdir, "two.h5"),
"dataframe",
format="table")
with dask.config.set(scheduler="sync"):
input_files = [
os.path.join(tdir, "one.h5"),
os.path.join(tdir, "two.h5")
]
res = dd.read_hdf(input_files, "dataframe")
tm.assert_frame_equal(res.compute(), df)
def test_hdf_file_list():
pytest.importorskip('tables')
df = pd.DataFrame({
'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]
},
index=[1., 2., 3., 4.])
with tmpdir() as tdir:
df.iloc[:2].to_hdf(os.path.join(tdir, 'one.h5'),
'dataframe',
format='table')
df.iloc[2:].to_hdf(os.path.join(tdir, 'two.h5'),
'dataframe',
format='table')
with dask.set_options(get=dask.get):
input_files = [
os.path.join(tdir, 'one.h5'),
os.path.join(tdir, 'two.h5')
]
res = dd.read_hdf(input_files, 'dataframe')
tm.assert_frame_equal(res.compute(), df)
def test_to_textfiles_name_function_warn():
seq = [
"a",
"b",
"c",
"d",
"e",
"f",
"g",
"h",
"i",
"j",
"k",
"l",
"m",
"n",
"o",
"p",
]
a = db.from_sequence(seq, npartitions=16)
with tmpdir() as dn:
with pytest.warns(None):
a.to_textfiles(dn, name_function=str)
def test_to_hdf_link_optimizations():
"""testing dask link levels is correct by calculating the depth of the dask graph"""
pytest.importorskip('tables')
df16 = pd.DataFrame(
{
'x': [
'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l',
'm', 'n', 'o', 'p'
],
'y': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]
},
index=[
1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15.,
16.
])
a = dd.from_pandas(df16, 16)
# saving to multiple hdf files, no links are needed
# expected layers: from_pandas, to_hdf, list = depth of 3
with tmpdir() as dn:
fn = os.path.join(dn, 'data*')
d = a.to_hdf(fn, '/data', compute=False)
assert dependency_depth(d.dask) == 3
# saving to a single hdf file with multiple nodes
# all subsequent nodes depend on the first
# expected layers: from_pandas, first to_hdf(creates file+node), subsequent to_hdfs, list = 4
with tmpfile() as fn:
d = a.to_hdf(fn, '/data*', compute=False)
assert dependency_depth(d.dask) == 4
# saving to a single hdf file with a single node
# every node depends on the previous node
# expected layers: from_pandas, to_hdf times npartitions(15), list = 2 + npartitions = 17
with tmpfile() as fn:
d = a.to_hdf(fn, '/data', compute=False)
assert dependency_depth(d.dask) == 2 + a.npartitions
