Environment Setup for Deep Learning

GT Big Data Bootcamp training material

Framework

PyTorch

We will use PyTorch on Python 3.6 as our main deep learning framework for the lab sessions for now. Later, we may expand these lab sessions with other popular deep learning frameworks such as TensorFlow and Theano. Your helps for adopting other frameworks or for enriching the lab materials are always welcome!

Jupyter Notebook

We will provide a Jupyter (iPython) Notebook file to practice some examples for each part of tutorials.

Environment

JupyterHub on Azure

We prepared a server with a GPU on Microsoft Azure. It would be enough to follow just the lab sessions even though the server is not very powerful one.

Access to the server

You can directly access to the JupyterHub with the pre-copied Notebooks used in the entire lab sessions on your web browser. Please open your favorite web browser and move to the following address including the port number:

http://52.175.231.74:8000

Then, you will meet the following login page. login

Please use guest for both Username and Password. Once you successfully login with the guest account, you can find the folders jupyter

  • Deep Learning Labs
    • Notebook files used in the tutorial.
  • Scratch Pads
    • Temporary folder for guests.

You can find the same notebook files used in the entire tutorials in Deep Learning Labs folder. Also, you can use Scratch Pads folder as your temporary storage. Please create a subfolder with a name YOUR-GT-ACCOUNT under Scratch Pads if you want to create a new notebook file for your own practice. Please note that we could clean up Scratch Pads folder occasionally without any notice.

Native

You can also use your native local machine as your environment if you want to.

Python Backend

We recommend you to use Anaconda for your Python backend (the tutorial notebooks are based on Python 3.6).

GPU acceleration

If you have a proper NVIDIA GPU(s) and want to utilize it, install CUDA Toolkit (7.5 or 8.0) including cuDNN before installing PyTorch.

Install PyTorch

Linux

CUDA 8.0
conda install pytorch torchvision cuda80 -c soumith
CUDA 7.5
conda install pytorch torchvision -c soumith

Mac (OSX)

CPU Only
conda install pytorch torchvision -c soumith

Mac users who want to use your GPU, you will need to build PyTorch from the source. Here is a good blog post about it (link).

Windows

There is no official support for Windows yet, but for Anaconda3 on Windows x64 (Windows 10, Windows Server 2016) you can try:

conda install -c peterjc123 pytorch

If you have some troubles, please refer to this pre-official discussion. It seems it will be merged into the official version soon!

Notebooks

If you want to download all notebook files to your local:

git clone https://github.com/ast0414/CSE6250BDH-LAB-DL.git

For more details, please refer to the official homepage of PyTorch.