MXNet Book

Hello, everyone.

I represent Packt Publishing in the acquisition of new material to publish. I have been working with one of their Product Managers, who may have reached out to you, and he asked for my help in securing an author or three for a book about MXNet.

It would be between the authors and Packt as to what, exactly, the book would entail. We wouldn’t want too much overlap between it and D2L, so I would imagine something more of a reference or cookbook.

I’d like to get the community’s thoughts on this. What kind of book would you like to see? What would you want a beginner to MXNet to know most about? What do you think would most help the MXNet community? And, most importantly, would you be willing to step up and help write this book?

Thank you for your time,
Matthew Emerick

I would not speak for everyone, but when I was beginner, the most problematic part was to create custom network from the ones that were given in the Symbolic API tutorials (those in tutorials are nice, but real use always includes customization). There is lack of short and clear explanation of Module, NDArrayIter classes, and data and label variables names. E.g. when you take network from tutorial and try to change output from SoftmaxOutput to SVMOutput, you start getting name strange mismatch labels error. Data layout would be good to explain too.
Clear tutorials with building custom networks, including ones with multiple input and output variables (like GANs, DAEs) would be nice.

1 Like

as EmilPi mentioned , i am very new and newbie on mxnet. most important problem for me is lack of tutorials and samples on mxnet modules and the guide to use toolkits .
1- modules
2- autograd guide if its still not deprecated
3- toolkits
4- loss functions and callbacks and other features
5- preprocessing guide and dataloaders and iterators
as i said im very newbie and my list maybe be lacking important topics