This structure completely describes a "receptive field" (a local linear model). More...
#include <lwpr.h>
Data Fields | |
| int | nReg |
| The number of PLS regression directions. | |
| int | nRegStore |
| The number of PLS directions that can be stored before a re-allocation is necessary. | |
| double * | fixStorage |
| A pointer to memory that is independent of nReg. | |
| double * | varStorage |
| A pointer to memory that might have to be re-allocated (nReg). | |
| int | trustworthy |
| This flag indicates whether a receptive field has "seen" enough data so that its predictions can be trusted. | |
| int | slopeReady |
| Indicates whether the vector "slope" can be used instead of doing PLS calculatations. | |
| double | w |
| The current activation (weight). | |
| double | sum_e2 |
| The accumulated prediction error on the training data. | |
| double | beta0 |
| Constant part of the PLS output. | |
| double | SSp |
| Sufficient statistics used for the confidence bounds. | |
| double * | D |
| Distance metric (NxN). | |
| double * | M |
| Cholesky factorization of the distance metric (NxN). | |
| double * | alpha |
| Learning rates for updates to M (NxN). | |
| double * | beta |
| PLS regression coefficients (Rx1). | |
| double * | c |
| The centre of the receptive field (Nx1). | |
| double * | SXresYres |
| Sufficient statistics for the PLS regression axes LWPR_ReceptiveField.U (NxR). | |
| double * | SSs2 |
| Sufficient statistics for PLS loadings s (Rx1). | |
| double * | SSYres |
| Sufficient statistics for PLS coefficients beta (Rx1). | |
| double * | SSXres |
| Sufficient statistics for PLS input reductions P (NxR). | |
| double * | U |
| PLS regression axes (NxR). | |
| double * | P |
| PLS input reduction parameters (NxR). | |
| double * | H |
| Sufficient statistics for distance metric updates (Rx1). | |
| double * | r |
| Sufficient statistics for distance metric updates (Rx1). | |
| double * | h |
| Sufficient statistics for 2nd order distance metric updates (NxN). | |
| double * | b |
| Memory terms for 2nd order updates to M (NxN). | |
| double * | sum_w |
| Accumulated activation w per PLS direction (Rx1). | |
| double * | sum_e_cv2 |
| Accumulated CV-error on training data (Rx1). | |
| double * | n_data |
| Number of training data each PLS direction has seen (Rx1). | |
| double * | lambda |
| Forgetting factor per PLS direction (Rx1). | |
| double * | mean_x |
| Mean of the training data this RF has seen (Nx1). | |
| double * | var_x |
| Variance of the training data this RF has seen (Nx1). | |
| double * | s |
| Current PLS loadings (Rx1). | |
| double * | slope |
| Slope of the local model (Nx1). This avoids PLS calculations when no updates are performed anymore. | |
| struct LWPR_Model * | model |
| Pointer to the LWPR_Model this RF belongs to. | |
This structure completely describes a "receptive field" (a local linear model).
In the descriptions of matrix- and vector-valued members of this structure, N denotes the input dimensionality, and R denotes the number of partial least squares (PLS) dimensions.
1.6.3