Machine Learning Workshop

Machine learning workshop given at the 2017 IGGI conference.

Data

Download the data from here. The files dataset_c4*.pkl are Connect Four data files generated by players of different strengths. You only need one of them -- I recomment dataset_c4.pkl.

The files are compressed to save space, but what they contain is essentially a list of dictionaries that looks something like this:

[
  {
    "own_pieces": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
    "opponent_pieces": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
    "move": 4,
    "own_win": 0,
    "opponent_win": 1
  },
  ...
  {
    "own_pieces": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
    "opponent_pieces": [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0],
    "move": 3,
    "own_win": 1,
    "opponent_win": 0
  }
]

own_pieces is a list representing the board (flattened) and has a 1 in spaces where the current player has a piece, and 0 in all other spaces.

opponent_pieces is the same but for the opponent.

move is the move that was made by the current player. This move is not reflected in own_pieces and opponent_pieces. In Connect Four, there are seven possible moves assuming that none of the columns are filled. move is therefore an integer in the range [0,6]. The data provider changes this to a one-hot encoding, e.g. 3 becomes [0, 0, 0, 1, 0, 0, 0].

own_win is 1 if the current player went on to win the game, and 0 if not.

opponent_win is the same but for the opponent.

TensorFlow installation instructions

Installation instructions can be found in the tf_install/ directory, or on tensorflow.org.

Windows

Python 3 installer included in tf_install/python/. Then run the commands in windows.bat.

Mac OS

Install Python 3 if you don't have it already. The included script will set up a virtualenv virtual environment. Activate it by running

source ./tf/bin/activate

and then run the second script to install TensorFlow.

Linux

The included script will download and install Python 3 and set up a virtualenv virtual environment. Activate it by running

source ./tf/bin/activate