Walks the schedule from t back to 0. With scale = 0
the schedule is the identity and this returns its input unchanged.
Usage
denoise(x_t, model, schedule, t_start = NULL)
Arguments
- x_t
Noisy state to start from.
- model
Function (x, t) returning the model's
x_0 estimate.
- schedule
Schedule list from noise_schedule.
- t_start
Starting timestep; defaults to the last step.
Value
A list mirroring the Python RichResult payload.
Examples
s <- noise_schedule(T = 10, scale = 0)
x <- c(1, 2, 3)
identical(denoise(x, model = function(x, t) x, schedule = s)$x0, x) # scale 0: the identity
#> [1] TRUE
