The R Project SVN R-packages

/*
 * commom variables for the rpart routine
 *
 * Start with things that depend on R.h
 */

#include <stddef.h>
#include <string.h> // for memcpy
#include <R.h>
#include <Rinternals.h>

#ifdef ENABLE_NLS
#include <libintl.h>
#define _(String) dgettext ("rpart", String)
#else
#define _(String) (String)
#endif

/*
 * Memory defined with R_alloc is removed automatically
 *  That with "CALLOC" I have to remove myself.  Use the
 *  latter for objects that need to persist between the
 *  s_to_rp1 and s_to_rp2 calls
 */
#define ALLOC(a,b)  R_alloc(a,b)
#define CALLOC(a,b) R_chk_calloc((size_t)(a), b)
#define RPARTNA(a) ISNAN(a)

/* done with the R internals */
#define LEFT  (-1)              /*used for the variable "extra" in nodes */
#define RIGHT  1
#define MISSING 0

#ifdef MAINRP
#define EXTERN
#else
#define EXTERN extern
#endif

/* As a sop to S, I need to keep the total number of external symbols
 *  somewhat smaller.  So, pack most of them all into a structure.
 */
EXTERN struct {
    double complexity;
    double alpha;
    double iscale;              /* used to check improvement==0, with error */
    double **ydata;
    double **xdata;
    double *xtemp;
    double *wt;
    double **ytemp;
    double *wtemp;              /* temp vector of weights */
    double *lwt;
    double *rwt;                /*scratch double vectors, of length ncat */
    double *vcost;              /* variable costs */
    int *numcat;                /* variable type: 0=cont, 1+  =#categories */
    int **sorts;                /* matrix of sort indices */
    int n;                      /* total number of subjects  */
    int num_y;                  /* number of y variables */
    int nvar;                   /* number of predictors */
    int maxpri;
    int maxsur;                 /* max # of primary or surrogate splits to use */
    int usesurrogate;
    int num_unique_cp;
    int min_node;               /* minimum size for any terminal node */
    int min_split;              /*minimum size before we attempt a split */
    int num_resp;               /*length of the response vector */
    int sur_agree;              /*0 =  my style, 1=CART style */
    int maxnode;                /*controls the maximum depth of the tree */
    int *tempvec;               /*to be allocated by the mainline, of length n */
    int *which;
    int *csplit;
    int *left;
    int *right;
} rp;

EXTERN struct cptable *cptable_tail;
EXTERN int (*rp_init) ();       /*called to initialize a splitting function */
EXTERN void (*rp_choose) ();    /*set to the splitting function */
EXTERN void (*rp_eval) ();      /*set to the evaluation routine */
EXTERN double (*rp_error) ();   /*set to the prediction error routine */
EXTERN int nodesize;

/*
 * The user inputs his complexity parameter as a percentage. and the
 *   printout is also scaled in this way.  The book and the computations all
 *   have an easier time with absolute cp.  So complex = what the user
 *   typed and alpha = complex * (risk of top node) = what is used
 *   internally.
 * The variable 'complex' in node.h is also on the internal scale.
 *
 * Categorical variables must be coded as 1,2,3, ..., and there may be
 *  missing categories.  The upper limit is determined on the fly.
 */