This project is a small Python study project about gradient descent in two dimensions. It compares four optimization algorithms on several test functions and plots their paths on contour charts.
Implemented algorithms:
- Simple gradient descent
- Momentum
- Nesterov accelerated gradient
- Adam
Available test functions:
fghRosenbrockBealeBooth
main.py: entry point used to run the visualizationdescent_algorithms.py: optimization algorithmsfunction.py: test functions and finite-difference gradient approximationTP_GradientDescent.pdf: assignment statementRapport_Gradient_Roden_Siffert_Rodriguez.pdf: original report
- Python 3.10+ recommended
numpymatplotlib
Create a virtual environment if needed, then install dependencies:
python3 -m pip install -r requirements.txtLaunch the script from the repository root:
python3 main.pyThis opens a Matplotlib window with the contour plot and the descent paths.
Example with explicit parameters:
python3 main.py --function rosenbrock --x0 -1.0 --y0 1.0 --learning-rate 0.001 --iterations 2000 --momentum-rate 0.9The script is configured through command-line arguments:
--function:f,g,h,rosenbrock,beale,booth--size: plot domain half-width--x0,--y0: initial point--learning-rate: learning rate--iterations: number of iterations--momentum-rate: momentum factor used by Momentum and Nesterov
No compilation step is required.
