Samplers¶
Samplers implement different algorithms for generating samples from the learned distribution. They define how noise is removed during the sampling process.
Base Sampler¶
The base class for all samplers. It defines the interface that all sampler implementations must follow.
Main Methods:
- __call__(x_T, score_model, ...): Performs the sampling process from initial noise x_T.
Euler-Maruyama¶
Implements the Euler-Maruyama numerical method for solving stochastic differential equations.
Main Methods:
- __call__(x_T, score_model, ...): Performs sampling using the Euler-Maruyama method.
Predictor-Corrector¶
Combines a predictor step with a corrector step based on Langevin dynamics for improved sampling quality.
Main Methods:
- predictor_step(x_t, t_curr, ...): Performs a prediction step.
- corrector_step(x_t, t, ...): Performs a correction step.
- __call__(x_T, score_model, ...): Performs sampling using the predictor-corrector method.
ODE Probability Flow¶
A deterministic sampling method based on the probability flow ordinary differential equation.
Main Methods:
- __call__(x_T, score_model, ...): Performs sampling using the ODE probability flow method.
Exponential Integrator¶
An exponential integration scheme for solving stochastic differential equations with better stability properties.
Main Methods:
- __call__(x_T, score_model, ...): Performs sampling using the exponential integrator method.