Help Center> ModelArts> MoXing Developer Guide> Sample Code for Advanced Applications
View PDF

Sample Code for Advanced Applications

If you are familiar with common operations, the MoXing Framework API document, and common Python code, you can refer to this section to use advanced MoXing Framework functions.

Closing a File After File Reading Is Completed

When an OBS file is read, an HTTP connection is called to read the network stream. You need to close the file after the file is read. To prevent you from forgetting to close a file, you are advised to use the with statement. When the with statement exits, the close() function of the mox.file.File object is automatically called.

1
2
 
with mox.file.File('obs://bucket_name/obs_file.txt', 'r') as f:
  data = f.readlines()

Reading or Writing an OBS File Using pandas

  • Use pandas to read an OBS file. 
    1
2
3
4
    		 
    import pandas as pd
import moxing as mox
with mox.file.File("obs://bucket_name/b.txt", "r") as f:
  csv = pd.read_csv(f)
  • Use pandas to write an OBS file. 
    1
2
3
4
5
    		 
    import pandas as pd
import moxing as mox
df = pd.DataFrame({'col1': [1, 2], 'col2': [3, 4]})
with mox.file.File("obs://bucket_name/b.txt", "w") as f:
  df.to_csv(f)

Reading an Image Using a File Object

When OpenCV is used to open an image, the OBS path cannot be passed and the image must be read using a file object. The following code cannot read the image:

1
2
 
import cv2
cv2.imread('obs://bucket_name/xxx.jpg', cv2.IMREAD_COLOR)

Modify the code as follows:

1
2
3
4
 
import cv2
import numpy as np
import moxing as mox
img = cv2.imdecode(np.fromstring(mox.file.read('obs://bucket_name/xxx.jpg', binary=True), np.uint8), cv2.IMREAD_COLOR)

Using an Existing API to Implement an API That Is Not Supported by mox.file

To call an API that is not supported by mox.file, you can use an existing API to implement the API and overwrite the original API. For example, os.path.isfile is not supported by mox.file. After mox.file.shift('os', 'mox') is called, os.path.isfile still calls the native builtin function of Python. You can overwrite the function with the following code: 
1
2
3
4
5
6
7
8
9
 
import os
import moxing as mox

_origin_isfile = os.path.isfile

def _patch_isfile(path):
  return not mox.file.isdir(path)

setattr(os.path, 'isfile', _patch_isfile)

Reconstructing an API That Does Not Support OBS Paths to an API That Supports OBS Paths

In pandas, to_hdf and read_hdf used to read and write H5 files do not support OBS paths, nor do they support file objects to be entered. The following code may cause errors:

1
2
3
4
 
import pandas as pd
df = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]}, index=['a', 'b', 'c'])
df.to_hdf('obs://wolfros-net/hdftest.h5', key='df', mode='w')
pd.read_hdf('obs://wolfros-net/hdftest.h5')

The API compiled using the pandas source code is rewritten to support OBS paths.

  • Write H5 to OBS = Write H5 to the local cache + Upload the local cache to OBS + Delete the local cache
  • Read H5 from OBS = Download H5 to the local cache + Read the local cache + Delete the local cache

That is, write the following code at the beginning of the script to enable to_hdf and read_hdf to support OBS paths:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
 
import os
import moxing as mox
import pandas as pd
from pandas.io import pytables
from pandas.core.generic import NDFrame

to_hdf_origin = getattr(NDFrame, 'to_hdf')
read_hdf_origin = getattr(pytables, 'read_hdf')


def to_hdf_override(self, path_or_buf, key, **kwargs):
  tmp_dir = '/cache/hdf_tmp'
  file_name = os.path.basename(path_or_buf)
  mox.file.make_dirs(tmp_dir)
  local_file = os.path.join(tmp_dir, file_name)
  to_hdf_origin(self, local_file, key, **kwargs)
  mox.file.copy(local_file, path_or_buf)
  mox.file.remove(local_file)


def read_hdf_override(path_or_buf, key=None, mode='r', **kwargs):
  tmp_dir = '/cache/hdf_tmp'
  file_name = os.path.basename(path_or_buf)
  mox.file.make_dirs(tmp_dir)
  local_file = os.path.join(tmp_dir, file_name)
  mox.file.copy(path_or_buf, local_file)
  result = read_hdf_origin(local_file, key, mode, **kwargs)
  mox.file.remove(local_file)
  return result

setattr(NDFrame, 'to_hdf', to_hdf_override)
setattr(pytables, 'read_hdf', read_hdf_override)
setattr(pd, 'read_hdf', read_hdf_override)

Use MoXing to Enable h5py.File to Support OBS

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
 
import os
import h5py
import numpy as np
import moxing as mox

h5py_File_class = h5py.File

class OBSFile(h5py_File_class):
  def __init__(self, name, *args, **kwargs):
    self._tmp_name = None
    self._target_name = name
    if name.startswith('obs://'):
      self._tmp_name = name.replace('/', '_')
      if mox.file.exists(name):
        mox.file.copy(name, os.path.join('cache', 'h5py_tmp', self._tmp_name))
      name = self._tmp_name

    super(OBSFile, self).__init__(name, *args, **kwargs)

  def close(self):
    if self._tmp_name:
      mox.file.copy(self._tmp_name, self._target_name)

    super(OBSFile, self).close()


setattr(h5py, 'File', OBSFile)

arr = np.random.randn(1000)
with h5py.File('obs://bucket/random.hdf5', 'r') as f:
  f.create_dataset("default", data=arr)

with h5py.File('obs://bucket/random.hdf5', 'r') as f:
  print(f.require_dataset("default", dtype=np.float32, shape=(1000,)))