A fitted (or queryable-before-fitting-throws) Gaussian Process regressor with a squared-exponential kernel: k(x, x') = sigma_f^2 * exp(-||x - x'||^2 / (2 * length_scale^2)), plus additive observation noise (a small noise_variance is also standard GP practice purely as numerical "jitter" to keep the kernel matrix well-conditioned, independent of whether the underlying objective is actually noisy).
More...
#include <gaussian_process.hpp>
|
| struct | Posterior |
| | A posterior prediction: mean and variance at one query point. More...
|
| |
|
| | GaussianProcessRegressor (float sigma_f, float length_scale, float noise_variance) |
| |
| void | Fit (std::vector< std::vector< float > > inputs, std::vector< float > targets) |
| | Fits the GP to (inputs[i], targets[i]) observation pairs – computes and solves the training kernel matrix once; Predict (below) reuses this fit.
|
| |
| Posterior | Predict (const std::vector< float > &x) const |
| | Predicts the posterior mean/variance at x, given the data passed to Fit.
|
| |
A fitted (or queryable-before-fitting-throws) Gaussian Process regressor with a squared-exponential kernel: k(x, x') = sigma_f^2 * exp(-||x - x'||^2 / (2 * length_scale^2)), plus additive observation noise (a small noise_variance is also standard GP practice purely as numerical "jitter" to keep the kernel matrix well-conditioned, independent of whether the underlying objective is actually noisy).
◆ GaussianProcessRegressor()
| pulsatrix::GaussianProcessRegressor::GaussianProcessRegressor |
( |
float |
sigma_f, |
|
|
float |
length_scale, |
|
|
float |
noise_variance |
|
) |
| |
|
inline |
- Parameters
-
| sigma_f | Kernel signal variance (prior variance at any single point, before conditioning on data). |
| length_scale | Kernel length scale – how far apart two points must be before the kernel considers them roughly uncorrelated. |
| noise_variance | Added to the kernel matrix's diagonal. Must be non-negative. |
- Exceptions
-
| std::invalid_argument | if sigma_f <= 0, length_scale <= 0, or noise_variance < 0. |
◆ Fit()
| void pulsatrix::GaussianProcessRegressor::Fit |
( |
std::vector< std::vector< float > > |
inputs, |
|
|
std::vector< float > |
targets |
|
) |
| |
|
inline |
Fits the GP to (inputs[i], targets[i]) observation pairs – computes and solves the training kernel matrix once; Predict (below) reuses this fit.
- Exceptions
-
| std::invalid_argument | if inputs is empty, inputs.size() != targets.size(), or inputs' rows are not all the same dimension. |
| std::runtime_error | – propagated from SolveLinearSystem if the kernel matrix is near-singular (e.g. duplicate input points with noise_variance == 0). |
◆ Predict()
| Posterior pulsatrix::GaussianProcessRegressor::Predict |
( |
const std::vector< float > & |
x | ) |
const |
|
inline |
Predicts the posterior mean/variance at x, given the data passed to Fit.
- Exceptions
-
| std::runtime_error | if Fit has not been called yet. |
| std::invalid_argument | if x's dimension doesn't match the training inputs'. |
| std::runtime_error | – propagated from SolveLinearSystem if the kernel matrix is near-singular. |
- Note
- variance is clamped to >= 0 – floating-point rounding can otherwise make it a tiny negative number exactly at (or extremely near) a training point, which is not a meaningful negative variance, just accumulated rounding error.
The documentation for this class was generated from the following file:
1.9.8