Introduction

The repository contains all the code developed for the Numerical Optimization for Large Scale Problem 2024-2025 course's project at Politecnico of Torino.

The repository is divided in 2 different folders: num for the Variants of Newton method developed in Matlab, and stoc for the Reinforcement Learning part developed in Python.

Reports are available in the respective pdfs: report_numerical_optimization.pdf and report_stochastic_optimization.pdf

We kindly advice to first read the reports provided to get a general idea of what the scope of this project is.

How to run

Stochastic module (Python)

All the development have been done assuming a python version greater or equal to 3.11, previous versions are not tested or directly supported. The following steps will guide you on the set-up to run the module:

1. Create a virtual environment using the built-in tool venv or the module virtualenv (all the instructions will consider only venv)

   python -m venv venv
   source venv/bin/activate   # on macOs/Linux
   venv\Scripts\activate.bat  # on Windows
   

2. Install all the required libraries (important: before running the following command be sure to be inside the virtual environment, otherwise the libraries will be installed globally)

   pip install -r requirements.txt # run it in the project root directory
   

3. Modify the constants in stoc/main.py if needed and then run:

   python stoc/main.py
   

Numerical module (Matlab)

To run the module simply download the repo and run one among MNM_test.m or TNM_test.m for obtaining results of respectively Modified Newton Method (Newton Method with matrix correction) and Truncated Newton Method. These will call the respective methods, defined as functions in modifiedNM.m and truncatedNM.m in a problem of choice, which can be set by following comments at the start of the test files.