MatteoMagnini / kill-experiments-woa-2022

KILL - knowledge injection via lambda layer

KILL-experiments (WOA 2022)

KILL - Knowledge Injection via Lambda Layer experiments (WOA 2022).

Users

Users can replicate the results of KILL algorithm applied to the poker hand dataset available at https://archive.ics.uci.edu/ml/datasets/Poker+Hand.

Requirements

• python 3.9
• java 11 (for Antlr4 support)
• antlr4-python3-runtime 4.9.3
• tensorflow 2.7.0
• scikit-learn 1.0.2
• pandas 1.4.2
• numpy 1.22.3
• 2ppy 0.4.0
• psyki 0.1.18
• scipy 1.8.0
• setuptools 62.1.0
• matplotlib 3.5.1

Setup

You can execute a list of predefined commands by running: python -m setup.py commandName.

Experiments

Run python -m setup.py run_kill_experiment to start a set of experiments. Options are:

• -k [y]/n to use symbolic knowledge injection;
• -e [y]/n to use early stop condition.

The command will start the training of 30 neural networks and their test. Results will be stored in resources/results/ folder inside knowledge or classic subfolders.

Statistics

Run python -m setup.py get_statistics to print the statistics for a given experiment population. Option are:

• -f [knowledge]/classic to specify the experiment class.

The command will print several performance metrics values and generate a plot concerning the class accuracy distributions. The plot will be available in resources/results folder.