No description, website, or topics provided.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

PAISS Deep Reinforcement Learning Practical Session by Criteo AI LAb

The repo contains all the material needed for a 1,5 hour practical session on Deep Reinforcement Learning.

The idea is to practice basic techniques so as to solve toy problems and also introduce useful techniques and diagnostic tools that allow to ultimately solve harder problems.

Authors that developed this educational material are cited in the AUTHORS file.

Setup Instructions

  1. Install Miniconda (cross-platform Python distribution):

    • download & run installer from here
  2. Create environment for the session:

    $ conda create --name paiss_deeprl python=3.6 Keras==2.2.0 tensorflow matplotlib
  3. Activate environment:

    $ source activate paiss_deeprl
  4. Install jupyter

    $ pip install jupyter
  5. Gym RL toolkit dependencies

    • Linux (known to work on Ubuntu 16.04)
      $ sudo apt-get install cmake swig zlib1g-dev
    • Mac OSX
      $ brew install cmake swig
    • Windows
  6. Gym RL toolkit install

    $ git clone
    $ cd gym
    $ pip install -e .[all]
  7. Verify everything is working

    $ cd - && python3 -c "import keras, gym"
  8. Pull the notebook with exercises

    $ git clone    

Exercises (FOR STUDENTS)

To run the notebook & start experimenting:

$ jupyter notebook exercises.ipynb

Note that there is also a PyTorch version : see exercises_pytorch.ipynb. If you wish to use it you'll need to conda install pytorch torchvision -c pytorch.


$ jupyter nbconvert slides.ipynb --to slides --post serve