MXNet Notebooks

This repo contains various notebooks ranging from basic usages of MXNet to state-of-the-art deep learning applications.

How to use

Python

The python notebooks are written in Jupyter.

• View We can view the notebooks on either github or nbviewer. But note that the former may be failed to render a page, while the latter has delays to view the recent changes.

• Run We can run and modify these notebooks if both mxnet and jupyter are installed. Here is an example script to install all these packages on Ubuntu.

  If you have a AWS account, here is an easier way to run the notebooks:

  1. Launch a p2.xlarge instance by using AMI ami-6e5d6808 on Ireland (eu-west-1). The Deep Learning AMI v2.0 for Amazon Linux is designed to continue to provide a stable, secure, and high performance execution environment for deep learning applications running on Amazon EC2. Remember to open the TCP port 8888 in the security group.

  2. Once launch is succeed, setup the following variable with proper value

    export HOSTNAME=ec2-107-22-159-132.compute-1.amazonaws.com
    export PERM=~/Downloads/my.pem

  3. Now we should be able to ssh to the machine by

       chmod 400 $PERM
       ssh -i $PERM -L 8888:localhost:8888 ec2-user@HOSTNAME

     Here we forward the EC2 machine's 8888 port into localhost.

  4. Clone this repo on the EC2 machine and run jupyter

       sudo yum install -y graphviz
       sudo mkdir /efs
       sudo chown ec2-user:ec2-user /efs
       cd /efs
       git clone https://github.com/dmlc/mxnet-notebooks
       jupyter notebook

     Leave this ssh session open and connected while using the python notebooks.

  5. Now we are able to view and edit the notebooks on the browser using the URL: http://localhost:8888/tree/mxnet-notebooks/python/outline.ipynb

  6. Finally you may want to connect another ssh session and run the following command to keep track of GPU memory and core usage

     ssh -i $PERM ec2-user@HOSTNAME
     watch -n 1 nvidia-smi

How to develop

Some general guidelines

• A notebook covers a single concept or application
• Try to be as basic as possible. Put advanced usages at the end, and allow reader to skip it.
• Keep the cell outputs on the notebooks so that readers can see the results without running