rgpycrumbs.surfaces

Attributes

k_matrix_matern_grad_map
k_matrix_se_grad_map
k_matrix_imq_grad_map
k_matrix_rq_grad_map

Classes

FastTPS
FastMatern
GradientMatern
FastIMQ
GradientSE
GradientIMQ
GradientRQ

Functions

safe_cholesky_solve(K, y, noise_scalar[, jitter_steps])	Retries Cholesky with increasing jitter if it fails.
generic_negative_mll(K, y, noise_scalar)
_tps_kernel_matrix(x)
negative_mll_tps(log_params, x, y)
_tps_solve(x, y, sm)
_tps_predict(x_query, x_obs, w, v)
_matern_kernel_matrix(x, length_scale)
negative_mll_matern_std(log_params, x, y)
_matern_solve(x, y, sm, length_scale)
_matern_predict(x_query, x_obs, alpha, length_scale)
matern_kernel_elem(x1, x2[, length_scale])
full_covariance_matern(x1, x2, length_scale)
negative_mll_matern_grad(log_params, x, y_flat, D_plus_1)
_grad_matern_solve(x, y_full, noise_scalar, length_scale)
_grad_matern_predict(x_query, x_obs, alpha, length_scale)
_imq_kernel_matrix(x, epsilon)
negative_mll_imq_std(log_params, x, y)
_imq_solve(x, y, sm, epsilon)
_imq_predict(x_query, x_obs, alpha, epsilon)
se_kernel_elem(x1, x2[, length_scale])
full_covariance_se(x1, x2, length_scale)
negative_mll_se_grad(log_params, x, y_flat, D_plus_1)
_grad_se_solve(x, y_full, noise_scalar, length_scale)
_grad_se_predict(x_query, x_obs, alpha, length_scale)
imq_kernel_elem(x1, x2[, epsilon])
full_covariance_imq(x1, x2, epsilon)
negative_mll_imq_grad(log_params, x, y_flat, D_plus_1)
negative_mll_imq_map(log_params, init_eps, x, y_flat, ...)
_grad_imq_solve(x, y_full, noise_scalar, epsilon)
_grad_imq_predict(x_query, x_obs, alpha, epsilon)
rq_kernel_base(x1, x2, length_scale, alpha)	Standard RQ Kernel: (1 + r^2 / (2*alpha*l^2))^-alpha
rq_kernel_elem(x1, x2, params)	SYMMETRIC KERNEL TRICK: k_sym(x, x') = k(x, x') + k(swap(x), x')
full_covariance_rq(x1, x2, params)
negative_mll_rq_map(log_params, x, y_flat, D_plus_1)
_grad_rq_solve(x, y_full, noise_scalar, params)
_grad_rq_predict(x_query, x_obs, alpha, params)
get_surface_model(name)

Module Contents

rgpycrumbs.surfaces.safe_cholesky_solve(K, y, noise_scalar, jitter_steps=3)

Retries Cholesky with increasing jitter if it fails.

rgpycrumbs.surfaces.generic_negative_mll(K, y, noise_scalar)

rgpycrumbs.surfaces._tps_kernel_matrix(x)

rgpycrumbs.surfaces.negative_mll_tps(log_params, x, y)

rgpycrumbs.surfaces._tps_solve(x, y, sm)

rgpycrumbs.surfaces._tps_predict(x_query, x_obs, w, v)

class rgpycrumbs.surfaces.FastTPS(x_obs, y_obs, smoothing=0.001, optimize=True, **kwargs)

x_obs

y_obs

y_mean

__call__(x_query, chunk_size=500)

Batched prediction to prevent OOM errors on large grids.

rgpycrumbs.surfaces._matern_kernel_matrix(x, length_scale)

rgpycrumbs.surfaces.negative_mll_matern_std(log_params, x, y)

rgpycrumbs.surfaces._matern_solve(x, y, sm, length_scale)

rgpycrumbs.surfaces._matern_predict(x_query, x_obs, alpha, length_scale)

class rgpycrumbs.surfaces.FastMatern(x_obs, y_obs, smoothing=0.001, length_scale=None, optimize=True, **kwargs)

x_obs

y_obs

y_mean

alpha

__call__(x_query, chunk_size=500)

Batched prediction for Matern.

rgpycrumbs.surfaces.matern_kernel_elem(x1, x2, length_scale=1.0)

rgpycrumbs.surfaces.full_covariance_matern(x1, x2, length_scale)

rgpycrumbs.surfaces.k_matrix_matern_grad_map

rgpycrumbs.surfaces.negative_mll_matern_grad(log_params, x, y_flat, D_plus_1)

rgpycrumbs.surfaces._grad_matern_solve(x, y_full, noise_scalar, length_scale)

rgpycrumbs.surfaces._grad_matern_predict(x_query, x_obs, alpha, length_scale)

class rgpycrumbs.surfaces.GradientMatern(x, y, gradients=None, smoothing=0.0001, length_scale=None, optimize=True, **kwargs)

x

y_full

e_mean

y_flat

alpha

__call__(x_query, chunk_size=500)

rgpycrumbs.surfaces._imq_kernel_matrix(x, epsilon)

rgpycrumbs.surfaces.negative_mll_imq_std(log_params, x, y)

rgpycrumbs.surfaces._imq_solve(x, y, sm, epsilon)

rgpycrumbs.surfaces._imq_predict(x_query, x_obs, alpha, epsilon)

class rgpycrumbs.surfaces.FastIMQ(x_obs, y_obs, smoothing=0.001, length_scale=None, optimize=True, **kwargs)

x_obs

y_obs

y_mean

alpha

__call__(x_query, chunk_size=500)

rgpycrumbs.surfaces.se_kernel_elem(x1, x2, length_scale=1.0)

rgpycrumbs.surfaces.full_covariance_se(x1, x2, length_scale)

rgpycrumbs.surfaces.k_matrix_se_grad_map

rgpycrumbs.surfaces.negative_mll_se_grad(log_params, x, y_flat, D_plus_1)

rgpycrumbs.surfaces._grad_se_solve(x, y_full, noise_scalar, length_scale)

rgpycrumbs.surfaces._grad_se_predict(x_query, x_obs, alpha, length_scale)

class rgpycrumbs.surfaces.GradientSE(x, y, gradients=None, smoothing=0.0001, length_scale=None, optimize=True, **kwargs)

x

y_full

e_mean

y_flat

alpha

__call__(x_query, chunk_size=500)

rgpycrumbs.surfaces.imq_kernel_elem(x1, x2, epsilon=1.0)

rgpycrumbs.surfaces.full_covariance_imq(x1, x2, epsilon)

rgpycrumbs.surfaces.k_matrix_imq_grad_map

rgpycrumbs.surfaces.negative_mll_imq_grad(log_params, x, y_flat, D_plus_1)

rgpycrumbs.surfaces.negative_mll_imq_map(log_params, init_eps, x, y_flat, D_plus_1)

rgpycrumbs.surfaces._grad_imq_solve(x, y_full, noise_scalar, epsilon)

rgpycrumbs.surfaces._grad_imq_predict(x_query, x_obs, alpha, epsilon)

class rgpycrumbs.surfaces.GradientIMQ(x, y, gradients=None, smoothing=0.0001, length_scale=None, optimize=True, **kwargs)

x

y_full

e_mean

y_flat

alpha

__call__(x_query, chunk_size=500)

rgpycrumbs.surfaces.rq_kernel_base(x1, x2, length_scale, alpha)

Standard RQ Kernel: (1 + r^2 / (2*alpha*l^2))^-alpha

rgpycrumbs.surfaces.rq_kernel_elem(x1, x2, params)

SYMMETRIC KERNEL TRICK: k_sym(x, x’) = k(x, x’) + k(swap(x), x’) Enforces f(r, p) = f(p, r) globally.

rgpycrumbs.surfaces.full_covariance_rq(x1, x2, params)

rgpycrumbs.surfaces.k_matrix_rq_grad_map

rgpycrumbs.surfaces.negative_mll_rq_map(log_params, x, y_flat, D_plus_1)

rgpycrumbs.surfaces._grad_rq_solve(x, y_full, noise_scalar, params)

rgpycrumbs.surfaces._grad_rq_predict(x_query, x_obs, alpha, params)

class rgpycrumbs.surfaces.GradientRQ(x, y, gradients=None, smoothing=0.0001, length_scale=None, optimize=True, **kwargs)

x

y_full

e_mean

y_flat

params

alpha

__call__(x_query, chunk_size=500)

rgpycrumbs.surfaces.get_surface_model(name)