akid is a python package written for doing research in Neural Network (NN). It
also aims to be production ready by taking care of concurrency and
communication in distributed computing (which depends on the utilities provided
by PyTorch and Tensorflow). It
could be seen as a user friendly front-end to torch, or tensorflow, like
Keras. It grows out of the motivation to reuse my old
code, and in the long run, to explore alternative framework for building NN. It
supports two backends, i.e., Tensorflow and Pytorch. If combining with
GlusterFS, Docker and
Kubernetes, it is able to provide dynamic and elastic
scheduling, auto fault recovery and scalability (which is not to brag
the capability of
akid, since the features are not features of
features thanks to open source (and libre software), but to mention the
possibilities that they can be combined.).
See http://akid.readthedocs.io/en/latest/index.html for documentation. The document is dated, and has not been updated to include new changes e.g., the PyTorch backend. But the backbone design is the same, and main features are there.
NOTE: the PyTorch end support has been way ahead of Tensorflow support now ...