Modern Deep Network Toolkits for pyTorch (MDNC)

This is a pyTorch framework used for

• Creating specially designed networks or layers.
• Parallel Data pre- and post- processing as a powerful alternative of torch.utils.data.DataLoader.
• Callback-function based data visualizer as an alternative of seaborn.
• Web tools for downloading tarball-packed datasets from Github.
• Some modified third-party utilities.

Usage

Currently, this project is still under-development. We suggest to use the following steps to add the package as a sub-module in your git-project,

cd <your-project-folder>
git submodule add https://github.com/cainmagi/MDNC.git mdnc
git submodule update --init --recursive

After that, you could use the pacakge by

from mdnc import mdnc

If you want to update the sub-module to the newest version, please use

git submodule update --remote --recursive

Progress

Now we have such progress on the semi-product:

• optimizers
• modules
  • conv: Modern convolutional layers and networks.
  • resnet: Residual blocks and networks.
  • resnext: ResNeXt blocks and networks.
  • incept: Google inception blocks and networks.
  • densenet: Dense-net blocks and networks.
• models
• data
  • h5py: Wrapped HDF5 datasets saver and loader.
  • netcdf4: Wrapped NETCDF4 datasets saver and loader.
  • bcolz: Wrapped Bcolz datasets saver and loader.
  • text: Wrapped text-based datasets saver and loader (CSV, JSON, TXT).
  • preprocs: Useful pre- and post- processing tools for all data handles in this package.
  • webtools: Web tools for downloading tarball-packed datasets from Github.
• funcs
• utils
  • tools: Light-weighted recording parsing tools used during training or testing.
  • draw: Wrapped matplotlib drawing tools. Most of the utilities are designed as call-back based functions.
• contribs
  • torchsummary 🔗: Keras style model.summary() in pyTorch, with some bugs gotten fixed (modified) (MIT licensed).
  • tensorboard 🔗: Wrapped torch.utils.tensorboard, supporting context-style writer and tensorboard.log converted to h5py format (not modified).

Documentation

View the document via the following link:

https://cainmagi.github.io/MDNC/

Demos

To be built now...

Debug reports

Currently, this project has not been checked by compatibility tests. During the developing stage, we are using pyTorch 1.7.0+ and Python 3.6+.

To perform the compatibility test, just run

cd <root-of-this-repo>
python -m mdnc

The compatibility test is shown as below. The checked item means this package performs well in the specific enviroment.

Enviroment	Win	Linux
pyTorch 1.7.0, Python 3.8	✅
pyTorch 1.8.0, Python 3.8	✅
pyTorch 1.6.0, Python 3.7
pyTorch 1.4.0, Python 3.7
pyTorch 1.2.0, Python 3.6
pyTorch 1.0.0, Python 3.5

Update reports

0.1.6 (alpha) @ 8/6/2021

1. Support thread_type for data.h5py.H*Parser.
2. Fix a bug when GPU is absent for data.sequence.

0.1.5 @ 3/14/2021

1. Add DecoderNet to our standard module protocol.
2. Fix some bugs of data.h5py and data.preprocs.
3. Make draw.setFigure enhanced by contextlib.
4. Add a title in Readme.md.
5. Fix typos and bugs in data and modules.
6. Add properties nlayers, input_size for networks in modules.

0.1.2 @ 2/27/2021

1. Fix more feature problems in contribs.torchsummary.
2. Fix bugs and finish data.preprocs.
3. Add more features in data.webtools.

0.1.0 @ 2/26/2021

1. Create this project.
2. Add packages: contribs, data, modules, utils.
3. Finish modules.conv, modules.resnet.
4. Finish data.h5py, data.webtools.
5. Finish contribs.torchsummary.
6. Drop the plan for supporting contribs.tqdm, add utils.ContexWrapper as for instead.
7. Add testing function for data.webtools.DataChecker.