Refactor python op namespaces

[reminisce]

1. mx.nd/sym.contrib
2. mx.nd/sym.random
3. mx.nd/sym.linalg
4. In ndarray.py and symbol.py, define sum = op.sum

[yancz1989]

Hi, @reminisce , how could I develop a new operator outside the mxnet src directory, and still able to call it using python? I may need to implement a new operator but using your prebuilt binary packages, will that be possible?

[reminisce]

I think it's possible but requires a lot of efforts on setting up the build environment for your implementation, including necessary header files and linking libs. It would be much easier if you could build from MXNet source code than using a prebuilt lib if you want to develop an operator. Is building from source feasible to you?

[yancz1989]

yes, of course. I would do it that way. Just one question, if I want to use intel MKL, is it true that I just need to download the intel_mklml from https://github.com/dmlc/web-data/raw/master/mxnet/mklml-release/mklml_lnx_<MKL VERSION>.tgz and don't need to install entire mkl from INTEL?

…

On Thu, Dec 7, 2017 at 3:59 AM, reminisce ***@***.***> wrote: I think it's possible but requires a lot of efforts on setting up the build environment for your implementation, including necessary header files and linking libs. It would be much easier if you could build from MXNet source code than using a prebuilt lib if you want to develop an operator. Is building from source feasible to you? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEts3_6Y4IXnv0z0apmwyYrla_BdJXP_ks5s9vIQgaJpZM4O_hqo> .

-- Chengzhe Yan Department of Automation, Tsinghua Univ. Cellphone: 8615101118689 Email: yancz1989@gmail.com

[reminisce]

I have not used MKL before. According the documentation, https://github.com/apache/incubator-mxnet/blob/master/MKL_README.md, it looks like what you said.