NeuralODE

We apply Neural Network, specifically MLP to solve ODE, and compare this method with traditional numerical method.

Install

Clone

git clone ssh://git@focs.ji.sjtu.edu.cn:2222/tang_jingfan/NeuralODE.git

Create virtual environment

conda create -n neural_ode python=3.8
conda activate neural_ode
pip install -r requirements.txt

Run

chmod +x drive
./drive --help                                # get help
./drive 2.0 0.5 0.2                           # just run
./drive 2.0 0.5 0.2 --plot                    # plot
./drive 2.0 0.5 0.2 --plot --rate --stability # plot, compare the efficiency, show stability

Result

Outputs

You can see the results in dir outputs.

We choose 4 traditional methods

• Runge-Kutta 4
• Adaptive RK
• Adams-Bashforth
• Adams-Moulton

comparing to NN (Neural Network).

If --stability argument exists, then original data points and the perturbation data points will be stored in ./outputs/points. The data points file with _p suffix represent the perturbation version. Then the stability.py will read the date from these points, and show the stabilities.

Img

You can see the plot, efficiecncy comparison and stability results in dir img.

Team

• Sun Zhengjia
• Tang Jingfan
• Zuo Tianyou