MXNet Learning (1)---the most accessible deep learning open Source Library---installation and environment building

Installation Environment: Win 10 Professional Edition 64-bit + Visual Studio Community.

Record the process of installing configuration mxnet in a GPU-equipped environment. The process uses Mxnet release's pre-built package directly, without using CMake compilation itself. Online has a lot of their own compiled tutorials, the process is more cumbersome, the direct use of the release package for beginners more simple and convenient.

The reason for choosing mxnet is because I read the "Comparison of three open source libraries in Caffe, TensorFlow, Mxnet", which points out that mxnet is relatively the easiest open source library for deep learning, so choose this open Source Library as a beginner tool.

　　　　　　

• Official documents

Getting Started: http://mxnet.readthedocs.io/en/latest/get_started/

• Installation

2. Configuring the Mxnet pre-built package
   • Download Link:https://github.com/dmlc/mxnet/releases
   It is divided into GPU-supported versions and GPU-enabled versions, and my computer supports GPUs, so select GPU versions.
   • Unzip the pre-built package to a suitable location, such as E:\MXNet;
   • Running Setupenv.cmd, this step helps to add various system environment variables that can be checked to see if their system environment variables are correct after running
     • The unpacked package has a Readme document that can be used as a reference in the installation process.
4. Configure CUDNN
   • Download Link:https://developer.nvidia.com/cudnn (need to register as a developer to download)
     
   • Unzip the CUDNN package, copy the bin, include, and Lib folders under the directory to the E:\MXNet\3rdparty\cudnn directory, overwriting the original file
6. Install Python2 (version 2.7.11)/Python3 (version 3.5.1)
• The best Python3 and Python2 are installed, although it is found that the following runs by default are Python3
• :https://www.python.org/downloads/ Select the 64-bit installation package
• The pre-built package is 64-bit, so remember to install 64-bit python, choose Custom Install, and tick the add path to system variable option, without checking manually add the Python installation path to the environment variable. Just started accidentally installed 32-bit, the result of running an error
8. Installing the Mxnet python package
   • In the CMD mode, enter the directory E:\MXNet\pythonas follows:
   • Run the python setup.pyInstall and wait for the installation to complete. The installation process will download the NumPy module that installs Python, remember to check for errors, and if wrong, rerun.
     • I am in the installation of the time because the network is not good, several times the timeout has been interrupted.
   • Run the instance code in the Readme document to see if it is correct

Refer to other articles and follow up with the following tools, which you can use to leave a memo here:

1. OpenCV (version 3.1):OpenCV3.1 has been released, but only 64-bit version. Requires a 32-bit version of the recommended use of OpenCV3.0

Download Link:http://opencv.org/downloads.html

2. Openblas

BLAS (Basic Linear Algebra subprograms): Basic linear Algebra Sub-Library, is an application interface (API) standard, the simple point is that vectors, matrices multiply these operations.

Openblas is a Blas open source optimization library that supports multi-threaded CPUs.

See the extracted MXNet pre-built package under the 3rdparty folder has Openblas, should not download it yourself.

The individual download links are:https://sourceforge.net/projects/openblas/files/latest/download?source=files

Related documents:https://github.com/xianyi/OpenBLAS/wiki/Document

3. CMake Tools

Very powerful compiler automatic configuration tool, can be based on different platforms, different compilers, to generate the corresponding makefile or vcproj project.

Download Link:https://cmake.org/download/

Getting Started reference:http://www.cnblogs.com/sinojelly/archive/2010/05/22/1741337.html

MXNet Learning (1)---the most accessible deep learning open Source Library---installation and environment building