Fedlab

A huge version update of FedLab. We further upgrade the APIs to better craft the FL simulation tasks. The updates can be broadly summarized as follows:

- Better source code file structure.
- Better README and tutorials.
- Unified abstract class interface, supporting plug-in usage of trainer and handler.
- Reproductions of the baseline algorithm are well built in fedlab.contrib.
- Built-in dataset partition tools.
- More examples.
- ...

Fixed typing errors in the old python version.

**The new version refactored the code and provided a more convenient FL interface.**
Please see our docs and source codes.

New feature:
1. Support Dockerfile deployment;
2. Better code organization structure;
3. Better communication APIs:
- Move p2p communication APIs "send and recv" into class DistNetwork;
- Package accepts any-shape tensors now (Only vectorized tensors are accepted in the previous version);
- Add communication volume information;

New Features：

1. Three stage communication manager: setup, communication, shutdown.
2. More flexible torch data type check.
3. More data partition strategy.

New feature:

1. DataPartitioner. More data partition strategy.
2. ModelMaintainer. Better class inheritance.
3. Compressor. New communication compressor group. Currently only topk available, BUT more are coming.