Jacobian Benchmarking

Scope

This project is about a performance investigation on different possible Jacobian function implementations in sympy. The ultimate goal is to implement one of these functions in the sympy library. This would be very useful when linearizing equations of motion in dynamic sytems.

Setup

Clone the repository:

git clone https://github.com/ricdigi/jacobian_benchmark.git
cd jacobian_benchmark

Install dependencies:
```
pip install -r requirements.txt
```

Directory Structure

jacobian_benchmark/
│
├── implementations/
│   ├── __init__.py
│   ├── forward_jacobian_ric.py     # @ricdigi's implementation
│   ├── forward_jacobian_ric2.py     # @ricdigi's 2 implementation  
│   ├── forward_jacobian_sam.py     # @brocksam's implementation
│   ├── jacobian_classic.py         # Classic sympy jacobian
│   ├── jacobian_protosym.py        # protosym based jacobian
    └── jacobian_symengine.py       # symengine based jacobian   
│
├── tests/
│   ├── __init__.py                 
│   ├── test_implementations.py     # Test the output of the functions
│   ├── conftest.py                 # Configuration file for pytest
│ 
│   
├── benchmark/
│   ├── __init__.py        # Initialization file for benchmark package
│   ├── benchmark.py       # Main benchmark logic
│   ├── models.py          # Collection of dynamical systems models
│   ├── utils.py           # Utility functions for profiling
│
├── data/
│   ├── results_pendulum.json  # File for storing pendulum benchmark results
│   ├── results_bicycle.json  # File for storing bicycle benchmark results
│   
│
├── requirements.txt       # List of dependencies
├── README.md              # Project documentation
├── plotting.py            # Plotting pendulum benchmark results
└── main.py                # Main script to run profiling

Usage

Run the main script to benchmark the implementations:

python main.py

The benchmark at the moment tests the performance of the implementations for increasing dimensions of the dynamical system. The system used is the n_link_pendulum_on_cart from sympy models.py. Results will be stored in the data directory.