Group Optimizer (Preferences + ILP)

Tkinter desktop app that collects each person’s preferences and then computes an assignment into groups that maximizes “preferred people in the same group” using Integer Linear Programming (PuLP / CBC).

What it does

1. Setup screen: enter:

   • a comma-separated list of names
   • N = max group size
   • M = number of preferences each person selects

2. Preference collection: each person, one-by-one, selects exactly M people (search + double-click to add, remove if needed).

3. Optimization: solves a binary ILP to maximize the number of directed preferences that end up within the same group.

4. Results: shows the resulting groups in a final read-only window.

How the scoring works

• Each person picks a list of preferred people.
• If person A prefers B, that counts as 1 point if A and B are assigned to the same group.
• Preferences are directed (A→B is not the same as B→A).

Limitations / behavior

• Group sizes are constrained by N (upper bound). Some groups may have fewer members if the total number of people is not a multiple of N.
• Feasibility is always guaranteed (assignment with capacity constraints), but optimality depends on the solver run.
• Preferences are binary (no ranking/weights beyond 0/1).

Requirements

• Python 3.9+ recommended
• Tkinter (usually included with Python on Windows/macOS; on some Linux distros you may need to install it separately)
• PuLP

Install

pip install pulp

Run

Save the script as main.py and run:

python main.py

Advanced clarifications

Notes on the optimizer (PuLP / CBC)

This project uses PuLP with CBC as the solver:

• In a normal Python environment, it uses:

  • pulp.PULP_CBC_CMD(msg=False)

• When packaged as a frozen app (PyInstaller / auto-py-to-exe), it tries to locate a bundled CBC binary under:

  • .../_MEIPASS/pulp/solverdir/cbc/win/i64/cbc.exe and then uses:
  • pulp.COIN_CMD(path=cbc_path, msg=False)

This is why the code checks getattr(sys, "frozen", False) and sys._MEIPASS.

Packaging (Windows) overview

If you package with auto-py-to-exe / PyInstaller, you typically need to ensure that the CBC executable is included in the build output so that the frozen app can find it.

Common approach:

• Add PuLP’s solver directory (or specifically the CBC executable) as “additional files” so it ends up in the expected _MEIPASS path.

The exact configuration depends on your environment and the layout of your PuLP installation.

Project structure

Single-file script (no separate modules). Main components:

• SetupApp: initial configuration UI
• PreferenceApp: preference collection UI
• build_weight_matrix(...): converts preferences to an adjacency/weight matrix
• solve_partition(...): ILP model construction + solve + group extraction