Decision Making

Auriau Vincent, Belahcene Khaled, Mousseau Vincent

Table of Contents

• Repository Usage
• Tasks
• Deliverables
• Resources

Repository usage

1. Install git-lfs , it will be needed to download the data
2. Clone the repository preferably using ssh
3. Make sure that git-lfs downloaded the files in data/. With the command du -sh *the files use several Mo of memory.

The command

conda env create -f config/env.yml
conda activate cs_td
python evaluation.py

will be used for evaluation, with two other test datasets. Make sure that it works well.

Dataset

You can find the first dataset in data/dataset_4. It contains three files: X.npy, Y.npy and Z.npy. They are organised so that $X[i] \succeq_{Z[i]} Y[i]$. Which means that the i-th element of X has been preferred to the i-th element of Y by the cluster described by the i-th element of Z. Of course Z is provided only for you to check your solution and shouldn't be used for the modelization. You can also use an equivalent with more features and pairs provided in data/dataset_10.

The second dataset needs to be downloaded through the choice-learn package. This notebook provides a few indications.

Tasks

You are asked to:

• Write a Mixed-Integer Progamming model that would solve both the clustering and learning of a UTA model on each cluster
• Code this MIP inside the TwoClusterMIP class in python/model.py. It should work on the dataset_4 dataset.
• Explain and code a heuristic model that can work on the cars dataset. It should be done inside the HeuristicModel class.

Deliverables

You will present your results during an oral presentation organized the on Tuesday $13^{th}$ (from 1.30 pm) of February. Exact time will be communicated later. Along the presentation, we are waiting for:

• A report summarizing you results as well as your thought process or even non-working models if you consider it to be interesting.
• Your solution of the first assignement should be clearly written in this report. For clarity, you should clearly state variables, constraints and objective of the MIP.
• A well organized git repository with all the Python code of the presented results. A GitHub fork of the repository is preferred. Add some documentation directly in the code or in the report for better understanding. The code must be easily run for testing purposes.
• In particular the repository should contain your solutions in the class TwoClustersMIP and HeuristicModel in the models.py file. If you use additional libraries, add them inside the config/env.yml file. The command 'python evaluation.py' will be used to check your models, be sure that it works and that your code complies with it. The dataset used will be a new one, with the same standards as 'dataset_4'.

Resources

• Gurobi
• Example Jupyter Notebook
• UTA model