| from __future__ import annotations | |
| import numpy as np | |
| def true_derivative(states: np.ndarray) -> np.ndarray: | |
| angle = states[..., 0] | |
| momentum = states[..., 1] | |
| return np.stack([momentum, -np.sin(angle)], axis=-1).astype(np.float32) | |
| def true_energy(states: np.ndarray) -> np.ndarray: | |
| angle = states[..., 0] | |
| momentum = states[..., 1] | |
| return 0.5 * momentum**2 + 1 - np.cos(angle) | |
| def generate_states(samples: int, seed: int) -> tuple[np.ndarray, np.ndarray]: | |
| rng = np.random.default_rng(seed) | |
| states = np.column_stack( | |
| [ | |
| rng.uniform(-np.pi, np.pi, samples), | |
| rng.uniform(-2.0, 2.0, samples), | |
| ] | |
| ).astype(np.float32) | |
| return states, true_derivative(states) | |