\-------/

aiopening

opening up AI - unleash the fury :)

aiopening is a Python library for games and reinforcement learning (RL) algorithms with focus on deep-neural-nets and deep-RL.

who should use aiopening?

Answer: people who ...

• are interested in RL algorithm research
• are interested in applying RL algorithms to solve computer game related problems

Apex Unleash Project

great things are about to come

How to setup your environment to run experiments in aiopening on Win10

aiopening only depends on the python3 libs: tensorflow/tensorflow-gpu and deepmind's sonnet library. Here is how to get setup with aiopening:

1. Install Anaconda3 (Python 3.6 version) on your PC

https://www.anaconda.com/download/

2. Open an "Anaconda Prompt" from your Win start menu
3. Create a new env and call it "aiopening" by typing:

>conda create -n aiopening python=3.6 numpy

4. Switch into the newly created aiopening env:

>activate aiopening

5. Install pygame
   • download the Windows wheel from http://www.lfd.uci.edu/~gohlke/pythonlibs/ (pygame‑1.9.3‑cp36‑cp36m‑win_amd64.whl)
   • and install it into the still activated aiopening Anaconda env via:

>pip install [path to pygame‑1.9.3‑cp36‑cp36m‑win_amd64.whl]

6. If your PC has a Nvidia GPU, follow the tensorflow GPU installation procedure here first (skipping everything from "Install Anaconda" on: https://nitishmutha.github.io/tensorflow/2017/01/22/TensorFlow-with-gpu-for-windows.html

7. Install the tensorflow package into our still active aiopening Anaconda env via:

>conda install tensorflow-gpu

8. Test your new tensorflow-gpu installation:

>python
...
>>> import tensorflow as tf
>>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

Check the printout after running the last command for mentions of: "(/gpu:0) -> (device: 0, name: GeForce GTX 980 ..."
NOTE: The name of your GPU may vary depending on your Nvidia GPU type.

9. Manually install deepmind's sonnet library in your aiopening env
   dm-sonnet is based on tensorflow and makes it very easy to build simple sub modules (e.g. a ConvNet unit) and then plug-and-play these modules to create larger and more complex NN topologies.

• While still in your conda env named aiopening: do a cd [any directory of your choice]
• git clone http://github.com/deepmind/sonnet or download and unzip sonnet directly from github: https://github.com/deepmind/sonnet/
• Rename setup.py.temp into setup.py
• In setup.py, change the line project_name = '%%%PROJECT_NAME%%%' to project_name = 'dm-sonnet-gpu'
• Rename the BUILD file into BUILD_test
• Now try in the directory of the setup.py file: python setup.py install. This should do a basic Win10 install (no C++ code compiled).

10. WIP: Not done yet. More instructions to follow on how to install aiopening itself ...