Version support, forward/backward compatibility, and python dependencies

Discussion
1

Hi, I have a few questions using MXNet with Python.

  1. What are the MXNet versions which are officially supported in the long term? I used MXNet 1.2.1 before, but I cannot find the git branch or tag for that version any more.

  2. It seems backward compatibility is guaranteed, i.e. models, which are trained with earlier versions of MXNet, can be loaded and used with later versions. What about forward compatibility? Can models trained with later versions used by earlier versions of MXNet?

  3. Where can I find the required versions of MXNet Python dependencies, e.g. Numpy, OpenCV? Is this file the source of truth? https://github.com/apache/incubator-mxnet/blob/v1.3.x/docs/install/requirements.txt

Thanks!

2

You can find the MXNet 1.2.1 version here https://github.com/apache/incubator-mxnet/tree/1.2.1

MXNet’s CI system has a backward compatibility checker, but no forward compatibility checker. So I assume forward compatibility is not guaranteed.

You can find the dependencies in https://github.com/apache/incubator-mxnet/blob/v1.3.x/docs/install/requirements.txt or on the website https://mxnet.incubator.apache.org/versions/master/install/