def test_zarrdatastore_list(self):
## WRITE
assert not os.path.exists(os.path.join(self.data_dir, 'test.mdb'))
zarr_datastore = ZarrDataStore(os.path.join(self.data_dir, 'test.mdb'),
datastore_type=DatastoreType.DIRECTORY,
compression_type=CompressionType.LZMA)
recorder = Recorder(zarr_datastore)
for i in range(self.n_arrays):
array = self.arrays[i]
recorder.record(self.key, array)
recorder.close()
## END WRITE
## READ
zarr_datastore = ZarrDataStore(os.path.join(self.data_dir, 'test.mdb'),
datastore_type=DatastoreType.DIRECTORY,
compression_type=CompressionType.LZMA)
recorder = Recorder(zarr_datastore)
l = recorder.get_all(self.key)
l = np.array(l)
self.assertTrue((self.arrays == l).all())
recorder.close()
def main():
read_only = False
data_dir = os.path.expanduser('~/output/tmp/redis-test')
os.makedirs(data_dir, exist_ok=True)
key = 'train.what'
bs = 10
sts = 500 * 20
ns = 200
n_arrays = 100
# serialization = Serialization.PYARROW
serialization = Serialization.PICKLE
if not read_only:
arrays = np.random.rand(n_arrays, bs, sts, ns)
if os.path.exists(data_dir):
print("Removing existing dir")
shutil.rmtree(data_dir)
os.makedirs(data_dir, exist_ok=True)
with RedisServer(data_directory=data_dir, serialization=serialization):
## WRITE
redis_datastore = RedisDataStore(server_host='localhost')
recorder = Recorder(redis_datastore)
with Timer() as wt:
write_times = []
for i in range(n_arrays):
array = arrays[i]
with Timer() as st:
recorder.record(key, array)
print("%d: Storing took %.2fs" % (i, st.difftime))
write_times.append(st.difftime)
print("Mean write time was %.4fs (+/- %.4f)" % (np.mean(write_times), np.std(write_times)))
recorder.close()
print("Total write time was %.2fs" % wt.difftime)
## END WRITE
print("Dir size after write is %d MiB" % (int(get_size(data_dir)) / 1024 / 1024))
## READ
with RedisServer(data_directory=data_dir, serialization=serialization):
redis_datastore = RedisDataStore(server_host='localhost', data_directory=data_dir, serialization=serialization)
recorder = Recorder(redis_datastore)
with Timer() as rt:
l = recorder.get_all(key)
print("Reading took %.2fs" % rt.difftime)
with Timer() as rrt:
l = np.array(l)
print("Into array took %.2f" % rrt.difftime)
print("Mean is", np.mean(l), l.shape)
recorder.close()
def test_inmemorydatastore_list(self):
## WRITE
inmem_datastore = InMemoryDataStore()
recorder = Recorder(inmem_datastore)
for i in range(self.n_arrays):
array = self.arrays[i]
recorder.record(self.key, array)
recorder.close()
## END WRITE
## READ
# If the in-memory datastore is initialized, all data will be reset!
recorder = Recorder(inmem_datastore)
l = recorder.get_all(self.key)
l = np.array(l)
self.assertTrue((self.arrays == l).all())
recorder.close()
def test_redisdatastore_list(self):
with RedisServer(data_directory=self.data_dir):
## WRITE
redis_datastore = RedisDataStore(server_host='localhost')
recorder = Recorder(redis_datastore)
for i in range(self.n_arrays):
array = self.arrays[i]
recorder.record(self.key, array)
recorder.close()
## END WRITE
with RedisServer(data_directory=self.data_dir):
## READ
redis_datastore = RedisDataStore(server_host='localhost')
recorder = Recorder(redis_datastore)
l = recorder.get_all(self.key)
l = np.array(l)
self.assertTrue((self.arrays == l).all())
recorder.close()
def test_hdf5datastore_list(self):
## WRITE
self.file_pth = os.path.join(self.data_dir, 'data.h5')
hdf5_datastore = HDF5DataStore(self.file_pth)
recorder = Recorder(hdf5_datastore)
for i in range(self.n_arrays):
array = self.arrays[i]
recorder.record(self.key, array)
# if i == 0:
# hdf5_datastore.enable_swmr()
recorder.close()
## END WRITE
## READ
hdf5_datastore = HDF5DataStore(self.file_pth)
recorder = Recorder(hdf5_datastore)
l = recorder.get_all(self.key)
l = np.array(l)
self.assertTrue((self.arrays == l).all())
recorder.close()
def test_zarrdatastore_object(self):
## WRITE
assert not os.path.exists(os.path.join(self.data_dir, 'test.mdb'))
zarr_datastore = ZarrDataStore(os.path.join(self.data_dir, 'test.mdb'),
datastore_type=DatastoreType.DIRECTORY,
compression_type=CompressionType.LZMA)
recorder = Recorder(zarr_datastore)
test_list_1 = [
np.random.rand(3, 1),
np.random.rand(4, 1),
np.array([])
]
test_list_2 = [
np.random.rand(3, 1),
np.random.rand(4, 1),
np.array([])
]
test_list = [test_list_1, test_list_2]
recorder.record(self.key, test_list_1)
recorder.record(self.key, test_list_2)
recorder.close()
## END WRITE
## READ
zarr_datastore = ZarrDataStore(os.path.join(self.data_dir, 'test.mdb'),
datastore_type=DatastoreType.DIRECTORY,
compression_type=CompressionType.LZMA)
recorder = Recorder(zarr_datastore)
l = recorder.get_all(self.key)
for j in range(len(l)):
for i in range(3):
self.assertTrue((test_list[j][i] == l[j][i]).all())
recorder.close()
def test_zarrdatastore_list_scalar(self):
## WRITE
assert not os.path.exists(os.path.join(self.data_dir, 'test.mdb'))
zarr_datastore = ZarrDataStore(os.path.join(self.data_dir, 'test.mdb'),
datastore_type=DatastoreType.DIRECTORY,
compression_type=CompressionType.LZMA)
recorder = Recorder(zarr_datastore)
recorder.record(self.key, 10.)
recorder.record(self.key, 20.)
recorder.close()
## END WRITE
## READ
zarr_datastore = ZarrDataStore(os.path.join(self.data_dir, 'test.mdb'),
datastore_type=DatastoreType.DIRECTORY,
compression_type=CompressionType.LZMA)
recorder = Recorder(zarr_datastore)
l = recorder.get_all(self.key)
l = np.array(l)
self.assertTrue((np.array([10., 20.]) == l).all())
recorder.close()
def main():
read_only = False
data_dir = os.path.expanduser('~/output/tmp/zarr-test')
file_pth = os.path.join(data_dir, 'data.mdb')
key = 'train/what'
bs = 10
sts = 500 * 20
ns = 200
n_arrays = 100
chunk_size_mb = 0.1
if not read_only:
arrays = np.random.rand(n_arrays, bs, sts, ns)
os.makedirs(data_dir, exist_ok=True)
if os.path.exists(file_pth):
print("Removing existing dir")
shutil.rmtree(file_pth)
## WRITE
zarr_datastore = ZarrDataStore(file_pth,
desired_chunk_size_bytes=chunk_size_mb *
1024**2)
recorder = Recorder(zarr_datastore)
with Timer() as wt:
write_times = []
for i in range(n_arrays):
array = arrays[i]
with Timer() as st:
recorder.record(key, array)
print("%d: Storing took %.2fs" % (i, st.difftime))
write_times.append(st.difftime)
print("Mean write time was %.4fs (+/- %.4f)" %
(np.mean(write_times), np.std(write_times)))
recorder.close()
print("Total write time was %.2fs" % wt.difftime)
## END WRITE
print("Dir size after write is %d MiB" %
(int(get_size(file_pth)) / 1024 / 1024))
## READ
zarr_datastore = ZarrDataStore(file_pth,
desired_chunk_size_bytes=chunk_size_mb *
1024**2)
recorder = Recorder(zarr_datastore)
with Timer() as rt:
l = recorder.get_all(key)
print("Reading took %.2fs" % rt.difftime)
read_times = []
for i in range(20):
b = np.random.randint(bs)
st = np.random.randint(sts)
with Timer() as rrt:
ll = np.array(l[:(n_arrays // 2), b, st, :])
print("Into sub-array took %.4fs" % rrt.difftime)
read_times.append(rrt.difftime)
print("Into sub-array mean readtime was %.4fs (+/- %.4f)" %
(np.mean(read_times), np.std(read_times)))
with Timer() as rrt:
l = np.array(l)
print("Into array (Total read time) was %.2fs" % rrt.difftime)
print("Data mean is", np.mean(l), l.shape)
recorder.close()