deap-practice

Practice script to play with DEAP's eaSimple method.

Build requirements

This project was developed and tested on Python 3.8. Any version that supports the packages within requirements.txt should work though.

It is highly recommended to set up a virtual environment or use Anaconda to set up this project.

Usage

DISCLAIMER: For the time being this only works on mathematical functions that take in a single input and provide a single output, e.g. x**4 - 4*x**3.

After running pip install -r requirements.txt run python main.py to run the tournament.

Modifying the inputs

To configure the script, modify the default args located within main.py.

If you wish to provide your own function and data set to play with generation, do the following:

• Create a custom data set. Follow resources/x4-4x3.txt for an example on how to format your data.
• Update ACTUAL_FUNC so it uses your function (instead of the default 'x**4 - 4*x**3').
• Update VALIDATION_FILE to point to your custom data set.

Interpreting the results

Running main.py displays roughly the following output (see resources/example_output.txt for an example):

<LOG RESULTS; only if verbose flag is set (default is True)>

<WINNING FUNCTION>

<WINNING FUNCTION'S ALGEBRAIC FORM>

<WINNING FUNCTION'S SIMPLIFIED FORM>

<WINNING FUNCTION'S EXPANDED FORM>

• Log results: The results of each generation within the tournament.

• Winning function: The best function generated as a result of running deap.algorithms.eaSimple.

• Algebraic form: The winning function's default display is in a functional format (e.g. add(x, x) instead of x + x). The code is ran through make_algebraic (see src/make_algebraic.py) to do this.

• Simplified form: The algebraic form can get rather long, so this shows the algebraic form ran through sympy.simplify.

• Expanded form: The simplified form can sometimes be clean up a bit more (e.g. terms can be merged or rearranged). To do this, the simplified form is ran through sympy.expand.

Resources

• DEAP docs: https://deap.readthedocs.io/en/master/index.html
• SymPy docs: https://docs.sympy.org/latest/index.html
• NumPy docs (using 1.9 as of writing this): https://numpy.org/doc/

Future plans

• Finalize folder structure for project (e.g. src instead of lib seems weird)
• Support more than just deap.algorithms.eaSimple
• Use argparse to provide full-blown CLI functionality
• Resolve TODOs in code

---

Copyright © 2020 Joseph Greene [email protected] Released under [The MIT License] (http://opensource.org/licenses/MIT) Project located at https://github.com/joegreene/misc-python-stuff