Testing and Validation

The project includes automated tests to keep the numerical and machine-learning components reliable as the code evolves.

Run the test suite with:

pytest

Current Test Coverage

The current tests cover:

• analytical Black-Scholes pricing, including a known at-the-money reference case and vectorized inputs;

• synthetic dataset generation, including schema, parameter ranges, non-negative prices, moneyness, reproducibility, and CSV round trips;

• controlled noisy-target generation, including zero-noise consistency, reproducibility, non-negative noisy targets, and experiment output writing;

• Monte Carlo pricing, including output shape, non-negative estimates, reproducibility, and approximate convergence to the analytical Black-Scholes price;

• model construction and activation-function selection;

• reduced-scale Support Vector Regression benchmarking;

• reduced-scale noisy-target SVR benchmarking;

• a small end-to-end pipeline smoke test, which verifies that a minimal experiment can run successfully and write the expected metrics, figures, and model artifacts.

Testing Philosophy

The tests are intentionally split between numerical correctness and integration checks.

Numerical tests validate the building blocks of the project, while the pipeline smoke test verifies that data generation, training, evaluation, Monte Carlo benchmarking, and artifact persistence work together. The smoke test uses a small temporary experiment configuration, so it remains fast and does not write outputs into the project directories.