Options - Options - 5.3 English - 68552

The available Python options are detailed in the aoclda.factorization.KernelPCA() class constructor.

The following options can be set using da_options_set_?:

Table 4.29 Kernel PCA options#

Option Name	Type	Default	Description	Constraints
kernel	string	\(s=\) linear	Kernel function to use.	\(s=\) linear, poly, precomputed, rbf, or sigmoid.
coef0	real	\(r=1\)	Independent term for polynomial and sigmoid kernels.	There are no constraints on \(r\).
eigen solver	string	\(s=\) auto	Which method to use for computing the eigendecomposition of the kernel matrix	\(s=\) auto, or syevd.
copy data	string	\(s=\) yes	Whether or not to store a copy of the training data.	\(s=\) no, or yes.
remove zero eig	string	\(s=\) no	Whether to remove components whose eigenvalue is zero.	\(s=\) no, or yes.
gamma	real	\(r=-1\)	Kernel coefficient for rbf, poly, and sigmoid kernels.	There are no constraints on \(r\).
fit inverse transform	string	\(s=\) no	Whether to fit the inverse transform.	\(s=\) no, or yes.
degree	integer	\(i=3\)	Degree for the polynomial kernel.	\(1 \le i\)
alpha	real	\(r=1\)	Ridge regularization parameter for the inverse transform linear solve.	\(0 < r\)
n_components	integer	\(i=0\)	Number of kernel principal components to compute.	\(0 \le i\)
check data	string	\(s=\) no	Check input data for NaNs prior to performing computation.	\(s=\) no, or yes.
storage order	string	\(s=\) column-major	Whether data is supplied and returned in row- or column-major order.	\(s=\) c, column-major, f, fortran, or row-major.

If n_components is set to 0, all principal components with a non-zero eigenvalue are retained.

When fit inverse transform is set to "yes", an approximate inverse transform is fitted during compute using kernel ridge regression. This is performed at compute time rather than on the first call to inverse_transform, so it carries additional memory and compute overhead. This option should only be set when inverse_transform will be used. It is incompatible with kernel='precomputed'.

When remove zero eig is enabled, components corresponding to near-zero eigenvalues are discarded after compute. The resulting n_components may be smaller than the value requested; query da_handle_get_result_int with da_kernel_pca_n_components to obtain the number of components computed.

When copy data is set to "no", the user’s data pointer is retained rather than copied and must remain valid through all subsequent calls.

For the rbf, poly, and sigmoid kernels, when gamma is not set (or set to a value less than \(0\)) it defaults to \(1/n_{\mathrm{features}}\) and is resolved at compute time. The resolved value can be retrieved via da_handle_get_result_? with da_kernel_pca_gamma.