ml_chaos_system

MATH303 project on reservoir computing for chaotic systems (currently focused on the Lorenz system).

What is in this folder

• lorenz.py: reproducible Lorenz + reservoir computing experiment pipeline
• euler_method.py: simple Euler simulation for the Lorenz system
• outputs_final/: generated figures and metrics from the tuned run (created after running lorenz.py)

Features in lorenz.py

• Deterministic runs with a fixed random seed
• Lorenz data generation via scipy.integrate.solve_ivp
• Reservoir computing training with ridge regression
• Validation-based hyperparameter search
• Test forecasting metrics (including valid prediction time)
• Lyapunov exponent estimation from the autonomous reservoir
• Saved plots and JSON metrics for report writing

Requirements

Use a Python environment with:

• numpy
• scipy
• matplotlib

You can install them with:

pip install -r requirements.txt

Run the experiment

Example tuned run:

MPLBACKEND=Agg MPLCONFIGDIR=.mplconfig python lorenz.py --trials 12 --forecast-steps 4000 --out-dir outputs_final

If your default Python does not have scientific packages, use your Conda Python (example):

MPLBACKEND=Agg MPLCONFIGDIR=.mplconfig /Users/sandyuelin/miniconda3/bin/python lorenz.py --trials 12 --forecast-steps 4000 --out-dir outputs_final

Output files

The script writes:

• metrics.json: final experiment metrics and best hyperparameters
• trial_records.json: validation scores from the hyperparameter search
• timeseries_test_forecast.png
• phase_portrait_test.png
• return_map_test.png
• hyperparameter_search.png