intro-deep-learning

Deep learning is an exciting topic in artificial intelligence. It has a lot of mystique surrounding it that can make it hard to understand what it entails. The goal of this project is to aggregate the best resources and provide introductory deep learning tutorials in the form of Jupyter Notebooks. The content will be explained as simply as possible, and the intended audience is anyone who has prior programming experience and interest in machine learning. I'm a deep learning hobbyist, and I hope that by putting together through these notebooks, I can more deeply (pun intended) understand concepts along the way and help you learn as well.

Knowledge needed to get started

For these notebooks, you'll need to be able to understand Python. If you are unfamiliar with Python, I suggest starting with:

https://github.com/kuleshov/cs228-material/blob/master/tutorials/python/cs228-python-tutorial.ipynb

Local Installation Instructions

1. Install Python 2.7

2. Install pip

3. Simply download the requirements.txt file to your local computer and run:
   pip install -r requirements.txt

AWS Installation Instructions

1. Follow instructions in https://medium.com/@DJVJallday/a-how-to-on-deep-reinforcement-learning-setup-aws-with-keras-tensorflow-openai-gym-and-jupyter-88bc0cc67e02

NOTE: You can skip the testing of OpenAI in step 20. This is only needed for reinforcement learning.

2. After completing all the installation instructions, do the optional instructions noted on the page:

sudo apt-get install libcupti-dev # needed for tensorflow
sudo pip install -U tensorflow # Python 2.7 (for python3 instructions, see https://www.tensorflow.org/install/install_linux#InstallingNativePip)

3. Use pip list to show the packages installed on the system. Validate the install and test the version:

python -c "import tensorflow as tf; print(tf.__version__)"
# you should see version x.y.z

Serving models via Tensorflow Serving

For more detailed instructions, see https://www.tensorflow.org/serving/serving_basic

Below are the basic ones

1. Install docker

See https://docs.docker.com/install/

Start up the application once it's installed.

2. Set up REST API hook

docker run -p 8501:8501 \
-v /Users/vrajaram/workspace/intro-deep-learning/saved_model_half_plus_three:/models/saved_model_half_plus_three \
-e MODEL_NAME=saved_model_half_plus_three -t tensorflow/serving &

3. Send request to the predict API

curl -d '{"instances": [1.0, 2.0, 5.0]}' -X POST http://localhost:8501/v1/models/saved_model_half_plus_three:predict

You should see:

{
  "predictions": [3.5, 4.0, 5.5]
}

See more at https://www.tensorflow.org/serving/docker

4. To kill the container do

docker kill $(docker ps -q)

To remove the images:

docker rmi $(docker images -q)

5. To save your own models and use in REST API, see: https://www.tensorflow.org/serving/serving_basic

Viewing notebooks online in nbviewer

If you do not want to use Github but rather Jupyter's nbviewer, visit https://nbviewer.jupyter.org/github/vr140/intro-deep-learning/tree/master/

Note that you can not actually run the notebooks here but rather just view them. To run them in the web, you'll need the Binder instructions section.

(Beta) Binder Instructions

Mybinder.org is a new initiative to allow people to run Jupyter notebooks on the web. A binder is set up at: https://mybinder.org/v2/gh/vr140/intro-deep-learning/master

NOTE: The binder is still in beta and may not work. For the most reliable environment, please see "Local Installation Instructions" above.