Personalized Algorithm Generation: A Case Study in Meta-Learning ODE Integrators

Requirements

This project uses python 3.8. Set up the project using the following step:

Install

$pip install -r requirements.txt

Look at examples.

Go to the folder
```
$cd examples
```
Run the training process (eg. linear traget)
```
$python main_linear.py
```
Change the parameters (such as learning rate, integrator stage, targeted order and so on) in configuration file
- For "linear target", a huge weight is multiplied on Taylor-based regularizer as a scale because this value is too small.
- For "square_nonlinear target", we multiply an increasing weight on MSE loss and a decreasing one on Taylor-based regularizer in order to focus on different loss in the training.

Ordercheck Run ipynb file ordercheck.
Plot the MSE on Van der Pol Oscillator and the Brussealtor. Run ipynb file plot error bar.
See the figures according to above evaluation in results.

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
.ipynb_checkpoints		.ipynb_checkpoints
examples		examples
intlib		intlib
results		results
README.md		README.md
ordercheck.ipynb		ordercheck.ipynb
plot_error_bar.ipynb		plot_error_bar.ipynb
requirements.txt		requirements.txt