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_hdf():
pytest.importorskip('tables')
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
with tmpfile('h5') as fn:
df.to_hdf(fn, '/data')
try:
dd.read_hdf(fn, 'data', chunksize=2)
assert False
except TypeError as e:
assert "format='table'" in str(e)
with tmpfile('h5') as fn:
df.to_hdf(fn, '/data', format='table')
a = dd.read_hdf(fn, '/data', chunksize=2)
assert a.npartitions == 2
assert a._known_dtype
tm.assert_frame_equal(a.compute(), df)
tm.assert_frame_equal(
dd.read_hdf(fn, '/data', chunksize=2, start=1, stop=3).compute(),
pd.read_hdf(fn, '/data', start=1, stop=3))
assert sorted(dd.read_hdf(fn, '/data').dask) == \
sorted(dd.read_hdf(fn, '/data').dask)
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_tokenize_numpy_memmap():
with tmpfile('.npy') as fn:
x = np.arange(5)
np.save(fn, x)
y = tokenize(np.load(fn, mmap_mode='r'))
with tmpfile('.npy') as fn:
x = np.arange(5)
np.save(fn, x)
z = tokenize(np.load(fn, mmap_mode='r'))
assert y != z
with tmpfile('.npy') as fn:
x = np.random.normal(size=(10, 10))
np.save(fn, x)
mm = np.load(fn, mmap_mode='r')
mm2 = np.load(fn, mmap_mode='r')
a = tokenize(mm[0, :])
b = tokenize(mm[1, :])
c = tokenize(mm[0:3, :])
d = tokenize(mm[:, 0])
assert len(set([a, b, c, d])) == 4
assert tokenize(mm) == tokenize(mm2)
assert tokenize(mm[1, :]) == tokenize(mm2[1, :])
def test_read_hdf(data, compare):
pytest.importorskip('tables')
with tmpfile('h5') as fn:
data.to_hdf(fn, '/data')
try:
dd.read_hdf(fn, 'data', chunksize=2, mode='r')
assert False
except TypeError as e:
assert "format='table'" in str(e)
with tmpfile('h5') as fn:
data.to_hdf(fn, '/data', format='table')
a = dd.read_hdf(fn, '/data', chunksize=2, mode='r')
assert a.npartitions == 2
compare(a.compute(), data)
compare(dd.read_hdf(fn, '/data', chunksize=2, start=1, stop=3,
mode='r').compute(),
pd.read_hdf(fn, '/data', start=1, stop=3))
assert (sorted(dd.read_hdf(fn, '/data', mode='r').dask) ==
sorted(dd.read_hdf(fn, '/data', mode='r').dask))
with tmpfile('h5') as fn:
sorted_data = data.sort_index()
sorted_data.to_hdf(fn, '/data', format='table')
a = dd.read_hdf(fn, '/data', chunksize=2, sorted_index=True, mode='r')
assert a.npartitions == 2
compare(a.compute(), sorted_data)
def test_to_hdf():
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)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data')
out = pd.read_hdf(fn, '/data')
tm.assert_frame_equal(df, out[:])
with tmpfile('h5') as fn:
a.x.to_hdf(fn, '/data')
out = pd.read_hdf(fn, '/data')
tm.assert_series_equal(df.x, out[:])
a = dd.from_pandas(df, 1)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data')
out = pd.read_hdf(fn, '/data')
tm.assert_frame_equal(df, out[:])
# test compute = False
with tmpfile('h5') as fn:
r = a.to_hdf(fn, '/data', compute=False)
r.compute()
out = pd.read_hdf(fn, '/data')
tm.assert_frame_equal(df, out[:])
def test_tokenize_numpy_memmap_no_filename():
# GH 1562:
with tmpfile('.npy') as fn1, tmpfile('.npy') as fn2:
x = np.arange(5)
np.save(fn1, x)
np.save(fn2, x)
a = np.load(fn1, mmap_mode='r')
b = a + a
assert tokenize(b) == tokenize(b)
def test_tokenize_numpy_memmap():
with tmpfile('.npy') as fn:
x = np.arange(5)
np.save(fn, x)
y = tokenize(np.load(fn, mmap_mode='r'))
with tmpfile('.npy') as fn:
x = np.arange(5)
np.save(fn, x)
z = tokenize(np.load(fn, mmap_mode='r'))
assert y != z
def test_to_csv_series():
s = pd.Series([1, 2, 3], index=[10, 20, 30], name="foo")
a = dd.from_pandas(s, 2)
with tmpfile("csv") as fn:
with tmpfile("csv") as fn2:
a.to_csv(fn)
s.to_csv(fn2)
with open(fn) as f:
adata = f.read()
with open(fn2) as f:
sdata = f.read()
assert adata == sdata
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]).compute()
assert os.path.exists(a)
assert os.path.exists(b)
with tmpfile() as dirname:
B.to_textfiles(dirname).compute()
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_memmap():
with tmpfile('npy') as fn_1:
with tmpfile('npy') as fn_2:
x = da.arange(100, chunks=15)
target = np.memmap(fn_1, shape=x.shape, mode='w+', dtype=x.dtype)
x.store(target)
assert eq(target, x)
np.save(fn_2, target)
assert eq(np.load(fn_2, mmap_mode='r'), x)
def test_ensure_file_directory(mkdir):
a = {'x': 1, 'y': {'a': 1}}
with tmpfile(extension='yaml') as source:
with tmpfile() as destination:
if mkdir:
os.mkdir(destination)
with open(source, 'w') as f:
yaml.dump(a, f)
ensure_file(source=source, destination=destination)
assert os.path.isdir(destination)
[fn] = os.listdir(destination)
assert os.path.split(fn)[1] == os.path.split(source)[1]
def test_to_hdf():
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, 2)
with tmpfile("h5") as fn:
a.to_hdf(fn, "/data")
out = pd.read_hdf(fn, "/data")
tm.assert_frame_equal(df, out[:])
with tmpfile("h5") as fn:
a.x.to_hdf(fn, "/data")
out = pd.read_hdf(fn, "/data")
tm.assert_series_equal(df.x, out[:])
def test_to_hdf_kwargs():
pytest.importorskip('tables')
df = pd.DataFrame({'A': ['a', 'aaaa']})
ddf = dd.from_pandas(df, npartitions=2)
with tmpfile('h5') as fn:
ddf.to_hdf(fn, 'foo4', format='table', min_itemsize=4)
df2 = pd.read_hdf(fn, 'foo4')
tm.assert_frame_equal(df, df2)
# test shorthand 't' for table
with tmpfile('h5') as fn:
ddf.to_hdf(fn, 'foo4', format='t', min_itemsize=4)
df2 = pd.read_hdf(fn, 'foo4')
tm.assert_frame_equal(df, df2)
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'))
with pytest.raises(TypeError):
B.to_textfiles(5)
def test_read_json_error():
with tmpfile('json') as f:
with pytest.raises(ValueError):
df.to_json(f, orient='split', lines=True)
df.to_json(f, orient='split', lines=False)
with pytest.raises(ValueError):
dd.read_json(f, orient='split', blocksize=1)
def test_gh715():
bin_data = u'\u20ac'.encode('utf-8')
with tmpfile() as fn:
with open(fn, 'wb') as f:
f.write(bin_data)
a = db.from_filenames(fn)
assert a.compute()[0] == bin_data.decode('utf-8')
def test_to_textfiles_endlines():
b = db.from_sequence(['a', 'b', 'c'], npartitions=1)
with tmpfile() as fn:
b.to_textfiles([fn])
with open(fn, 'r') as f:
result = f.readlines()
assert result == ['a\n', 'b\n', 'c']
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_gh606():
encoding = 'utf-16-le'
euro = u'\u20ac'
yen = u'\u00a5'
linesep = os.linesep
bin_euro = u'\u20ac'.encode(encoding)
bin_yen = u'\u00a5'.encode(encoding)
bin_linesep = linesep.encode(encoding)
data = (euro * 10) + linesep + (yen * 10) + linesep + (euro * 10)
bin_data = data.encode(encoding)
with tmpfile() as fn:
with open(fn, 'w+b') as f:
f.write(bin_data)
f.seek(0)
stop = len(bin_euro) * 10 + len(bin_linesep)
res = textblock(f, 1, stop, encoding=encoding)
assert res == ((yen * 10) + linesep).encode(encoding)
stop = len(bin_euro) * 10 + len(bin_linesep)
res = textblock(f, 0, stop, encoding=encoding)
assert res == ((euro * 10) + linesep + (yen * 10) + linesep).encode(encoding)
def test_nextlinesep():
lineseps = ('\r', '\n', '\r\n')
encodings = ('utf-16-le', 'utf-8')
for sep, encoding in product(lineseps, encodings):
euro = u'\u20ac'
yen = u'\u00a5'
bin_euro = u'\u20ac'.encode(encoding)
bin_yen = u'\u00a5'.encode(encoding)
bin_sep = sep.encode(encoding)
data = (euro * 10) + sep + (yen * 10) + sep + (euro * 10)
bin_data = data.encode(encoding)
with tmpfile() as fn:
with open(fn, 'w+b') as f:
f.write(bin_data)
f.seek(0)
start, stop = next_linesep(f, 5, encoding, sep)
assert start == len(bin_euro) * 10
assert stop == len(bin_euro) * 10 + len(sep.encode(encoding))
seek = len(bin_euro) * 10 + len(bin_sep) + len(bin_yen)
start, stop = next_linesep(f, seek, encoding, sep)
exp_start = len(bin_euro) * 10 + len(bin_sep) + len(bin_yen) * 10
exp_stop = exp_start + len(bin_sep)
assert start == exp_start
assert stop == exp_stop
def test_linecount_gzip():
with tmpfile('gz') as fn:
f = gzip.open(fn, 'wb')
for line in text.split('\n'):
f.write(line.encode('ascii'))
f.write(b'\n')
f.close()
assert linecount(fn) == 7
def test_to_textfiles_endlines():
b = db.from_sequence(['a', 'b', 'c'], npartitions=1)
with tmpfile() as fn:
for last_endline in False, True:
b.to_textfiles([fn], last_endline=last_endline)
with open(fn, 'r') as f:
result = f.readlines()
assert result == ['a\n', 'b\n', 'c\n' if last_endline else 'c']
def test_to_csv():
df = pd.DataFrame({'x': ['a', 'b', 'c', 'd'],
'y': [1, 2, 3, 4]}, index=[1., 2., 3., 4.])
for npartitions in [1, 2]:
a = dd.from_pandas(df, npartitions)
with tmpfile('csv') as fn:
a.to_csv(fn)
result = pd.read_csv(fn, index_col=0)
tm.assert_frame_equal(result, df)
a = dd.from_pandas(df, npartitions)
with tmpfile('csv') as fn:
r = a.to_csv(fn, compute=False)
r.compute()
result = pd.read_csv(fn, index_col=0)
tm.assert_frame_equal(result, df)
def test_linecount_bz2():
with tmpfile('bz2') as fn:
f = bz2.BZ2File(fn, 'wb')
for line in text.split('\n'):
f.write(line.encode('ascii'))
f.write(b'\n')
f.close()
assert linecount(fn) == 7
def test_visualize_order():
pytest.importorskip('matplotlib')
x = da.arange(5, chunks=2)
with tmpfile(extension='dot') as fn:
x.visualize(color='order', filename=fn, cmap='RdBu')
with open(fn) as f:
text = f.read()
assert 'color="#' in text
def test_to_hdf_modes_multiple_nodes():
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 tmpfile('h5') as fn:
a.to_hdf(fn, '/data2')
a.to_hdf(fn, '/data*', mode='a')
out = dd.read_hdf(fn, '/data*')
eq(df.append(df), out)
# overwriting a file with a single partition
a = dd.from_pandas(df, 1)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data2')
a.to_hdf(fn, '/data*', mode='w')
out = dd.read_hdf(fn, '/data*')
eq(df, out)
# appending two partitions to existing data
a = dd.from_pandas(df, 2)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data2')
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 tmpfile('h5') as fn:
a.to_hdf(fn, '/data2')
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 tmpfile('h5') as fn:
a.to_hdf(fn, '/data1')
a.to_hdf(fn, '/data2')
a.to_hdf(fn, '/data*', mode='a', append=False)
out = dd.read_hdf(fn, '/data*')
eq(df.append(df), out)
def test_read_hdf_doesnt_segfault():
with tmpfile("h5") as fn:
N = 40
df = pd.DataFrame(np.random.randn(N, 3))
with pd.HDFStore(fn, mode="w") as store:
store.append("/x", df)
ddf = dd.read_hdf(fn, "/x", chunksize=2)
assert len(ddf) == N
def test_to_csv_gzip():
df = pd.DataFrame({"x": ["a", "b", "c", "d"], "y": [1, 2, 3, 4]}, index=[1.0, 2.0, 3.0, 4.0])
for npartitions in [1, 2]:
a = dd.from_pandas(df, npartitions)
with tmpfile("csv") as fn:
a.to_csv(fn, compression="gzip")
result = pd.read_csv(fn, index_col=0, compression="gzip")
tm.assert_frame_equal(result, df)
def test_from_bcolz_filename():
bcolz = pytest.importorskip("bcolz")
with tmpfile(".bcolz") as fn:
t = bcolz.ctable([[1, 2, 3], [1.0, 2.0, 3.0], ["a", "b", "a"]], names=["x", "y", "a"], rootdir=fn)
t.flush()
d = dd.from_bcolz(fn, chunksize=2)
assert list(d.x.compute()) == [1, 2, 3]
def test_to_hdf_multiple_datasets():
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 datasets making sure order is kept
with tmpfile('h5') as fn:
b.to_hdf(fn, '/data*')
out = dd.read_hdf(fn, '/data*')
eq(df16, out)
# saving to multiple datasets
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data*')
out = dd.read_hdf(fn, '/data*')
eq(df, out)
# 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')
eq(df, out)
# saving to multiple datasets with custom name_function
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data_*', name_function=lambda i: 'a' * (i + 1))
out = dd.read_hdf(fn, '/data_*')
eq(df, out)
out = pd.read_hdf(fn, '/data_a')
tm.assert_frame_equal(out, df.iloc[:2])
out = pd.read_hdf(fn, '/data_aa')
tm.assert_frame_equal(out, df.iloc[2:])
# 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')
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:])
# 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_*_*')
# 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*')
eq(df, out)
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=[
1.0,
2.0,
3.0,
4.0,
5.0,
6.0,
7.0,
8.0,
9.0,
10.0,
11.0,
12.0,
13.0,
14.0,
15.0,
16.0,
],
)
a = dd.from_pandas(df16, 16)
# adding artificial 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, meta=a)
a.to_hdf(fn, "/data*")
out = dd.read_hdf(fn, "/data*")
assert_eq(df16, out)
# saving to multiple hdf files
# adding artificial 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, meta=a)
a.to_hdf(fn, "/data")
out = dd.read_hdf(fn, "/data")
assert_eq(df16, out)
def test_to_hdf_schedulers(scheduler, npartitions):
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=[
1.0,
2.0,
3.0,
4.0,
5.0,
6.0,
7.0,
8.0,
9.0,
10.0,
11.0,
12.0,
13.0,
14.0,
15.0,
16.0,
],
)
a = dd.from_pandas(df, npartitions=npartitions)
# test single file single node
with tmpfile("h5") as fn:
a.to_hdf(fn, "/data", scheduler=scheduler)
out = pd.read_hdf(fn, "/data")
assert_eq(df, out)
# test multiple files single node
with tmpdir() as dn:
fn = os.path.join(dn, "data_*.h5")
a.to_hdf(fn, "/data", scheduler=scheduler)
out = dd.read_hdf(fn, "/data")
assert_eq(df, out)
# test single file multiple nodes
with tmpfile("h5") as fn:
a.to_hdf(fn, "/data*", scheduler=scheduler)
out = dd.read_hdf(fn, "/data*")
assert_eq(df, out)
def test_to_hdf_multiple_nodes():
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, 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=[
1.0,
2.0,
3.0,
4.0,
5.0,
6.0,
7.0,
8.0,
9.0,
10.0,
11.0,
12.0,
13.0,
14.0,
15.0,
16.0,
],
)
b = dd.from_pandas(df16, 16)
# saving to multiple nodes
with tmpfile("h5") as fn:
a.to_hdf(fn, "/data*")
out = dd.read_hdf(fn, "/data*")
assert_eq(df, out)
# saving to multiple nodes making sure order is kept
with tmpfile("h5") as fn:
b.to_hdf(fn, "/data*")
out = dd.read_hdf(fn, "/data*")
assert_eq(df16, out)
# saving to multiple datasets with custom name_function
with tmpfile("h5") as fn:
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(fn, "/data_a")
tm.assert_frame_equal(out, df.iloc[:2])
out = pd.read_hdf(fn, "/data_aa")
tm.assert_frame_equal(out, df.iloc[2:])
# test multiple nodes with hdf object
with tmpfile("h5") as fn:
with pd.HDFStore(fn) as hdf:
b.to_hdf(hdf, "/data*")
out = dd.read_hdf(fn, "/data*")
assert_eq(df16, out)
def db():
with tmpfile() as f:
uri = "sqlite:///%s" % f
df.to_sql("test", uri, index=True, if_exists="replace")
yield uri
def tmp_db_uri():
with tmpfile() as f:
yield "sqlite:///%s" % f
def test_roundtrip_from_pandas(engine):
with tmpfile() as fn:
df = pd.DataFrame({'x': [1, 2, 3]})
fastparquet.write(fn, df)
ddf = dd.io.parquet.read_parquet(fn, index=False, engine=engine)
assert_eq(df, ddf)
def test_to_hdf():
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)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data')
out = pd.read_hdf(fn, '/data')
tm.assert_frame_equal(df, out[:])
with tmpfile('h5') as fn:
a.x.to_hdf(fn, '/data')
out = pd.read_hdf(fn, '/data')
tm.assert_series_equal(df.x, out[:])
a = dd.from_pandas(df, 1)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data')
out = pd.read_hdf(fn, '/data')
tm.assert_frame_equal(df, out[:])
# saving to multiple datasets
a = dd.from_pandas(df, 2)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data*')
out = dd.read_hdf(fn, '/data*')
tm.assert_frame_equal(df, out.compute())
# saving to multiple files
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data')
out = dd.read_hdf(fn, '/data')
tm.assert_frame_equal(df, out.compute())
# saving to multiple datasets with custom name_function
a = dd.from_pandas(df, 2)
with tmpfile('h5') as fn:
a.to_hdf(fn, '/data_*', name_function=lambda i: 'a' * (i + 1))
out = dd.read_hdf(fn, '/data_*')
tm.assert_frame_equal(df, out.compute())
out = pd.read_hdf(fn, '/data_a')
tm.assert_frame_equal(out, df.iloc[:2])
out = pd.read_hdf(fn, '/data_aa')
tm.assert_frame_equal(out, df.iloc[2:])
# saving to multiple files with custom name_function
a = dd.from_pandas(df, 2)
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')
tm.assert_frame_equal(df, out.compute())
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:])
# saving to different datasets in multiple files with custom name_function
a = dd.from_pandas(df, 2)
with tmpdir() as dn:
with pytest.raises(ValueError):
fn = os.path.join(dn, 'data_*.h5')
a.to_hdf(fn, '/data_*', name_function=lambda i: 'a' * (i + 1))
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=[
1.0,
2.0,
3.0,
4.0,
5.0,
6.0,
7.0,
8.0,
9.0,
10.0,
11.0,
12.0,
13.0,
14.0,
15.0,
16.0,
],
)
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
