1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
|
/**
Random forest with conditional inference trees
*\file RandomForest.c
*\author $Author: thothorn $
*\date $Date: 2019-03-04 11:23:08 +0100 (Mo, 04 Mär 2019) $
*/
#include "party.h"
/**
An experimental implementation of random forest like algorithms \n
*\param learnsample an object of class `LearningSample'
*\param weights a vector of case weights
*\param bwhere integer matrix (n x ntree) for terminal node numbers
*\param bweights double matrix (n x ntree) for bootstrap case weights
*\param fitmem an object of class `TreeFitMemory'
*\param controls an object of class `TreeControl'
*/
SEXP R_Ensemble(SEXP learnsample, SEXP weights, SEXP controls) {
SEXP ans, nweights, tree, where, ensemble, bw, fitmem, bwhere, bweights;
double *dnweights, *dweights, sw = 0.0, *prob, tmp;
int nobs, i, b, B , nodenum = 1, *iweights, *iweightstmp,
*iwhere, replace, fraction, wgrzero = 0, realweights = 0;
int j, k, l, swi = 0;
B = get_ntree(controls);
nobs = get_nobs(learnsample);
PROTECT(ans = party_NEW_OBJECT("RandomForest"));
PROTECT(bwhere = allocVector(VECSXP, B));
PROTECT(bweights = allocVector(VECSXP, B));
PROTECT(ensemble = allocVector(VECSXP, B));
PROTECT(fitmem = ctree_memory(learnsample, PROTECT(ScalarLogical(1))));
SET_SLOT(ans, PL2_ensembleSym, ensemble);
SET_SLOT(ans, PL2_whereSym, bwhere);
SET_SLOT(ans, PL2_weightsSym, bweights);
iweights = Calloc(nobs, int);
iweightstmp = Calloc(nobs, int);
prob = Calloc(nobs, double);
dweights = REAL(weights);
for (i = 0; i < nobs; i++) {
/* sum of weights */
sw += dweights[i];
/* number of weights > 0 */
if (dweights[i] > 0) wgrzero++;
/* case weights or real weights? */
if (dweights[i] - ftrunc(dweights[i]) > 0)
realweights = 1;
}
for (i = 0; i < nobs; i++)
prob[i] = dweights[i]/sw;
swi = (int) ftrunc(sw);
replace = get_replace(controls);
/* fraction of number of obs with weight > 0 */
if (realweights) {
/* fraction of number of obs with weight > 0 for real weights*/
tmp = (get_fraction(controls) * wgrzero);
} else {
/* fraction of sum of weights for case weights */
tmp = (get_fraction(controls) * sw);
}
fraction = (int) ftrunc(tmp);
if (ftrunc(tmp) < tmp) fraction++;
if (!replace) {
if (fraction < 10)
error("fraction of %f is too small", fraction);
}
/* <FIXME> can we call those guys ONCE? what about the deeper
calls??? </FIXME> */
GetRNGstate();
if (get_trace(controls))
Rprintf("\n");
for (b = 0; b < B; b++) {
SET_VECTOR_ELT(ensemble, b, tree = allocVector(VECSXP, NODE_LENGTH + 1));
SET_VECTOR_ELT(bwhere, b, where = allocVector(INTSXP, nobs));
SET_VECTOR_ELT(bweights, b, bw = allocVector(REALSXP, nobs));
iwhere = INTEGER(where);
for (i = 0; i < nobs; i++) iwhere[i] = 0;
C_init_node(tree, nobs, get_ninputs(learnsample),
get_maxsurrogate(get_splitctrl(controls)),
ncol(get_predict_trafo(GET_SLOT(learnsample,
PL2_responsesSym))));
/* generate altered weights for perturbation */
if (replace) {
/* weights for a bootstrap sample */
rmultinom(swi, prob, nobs, iweights);
} else {
/* weights for sample splitting */
C_SampleSplitting(nobs, prob, iweights, fraction);
}
nweights = S3get_nodeweights(tree);
dnweights = REAL(nweights);
for (i = 0; i < nobs; i++) {
REAL(bw)[i] = (double) iweights[i];
dnweights[i] = REAL(bw)[i];
}
C_TreeGrow(tree, learnsample, fitmem, controls, iwhere, &nodenum, 1);
nodenum = 1;
C_remove_weights(tree, 0);
/* compute terminal node ids for all observations (not just those with weight > 0) */
for (i = 0; i < nobs; i++)
iwhere[i] = C_get_nodeID(tree, GET_SLOT(learnsample, PL2_inputsSym), 0.0, i, -1);
if (get_trace(controls)) {
/* progress bar; inspired by
http://avinashjoshi.co.in/2009/10/13/creating-a-progress-bar-in-c/ */
Rprintf("[");
/* Print the = until the current percentage */
l = (int) ceil( ((double) b * 50.0) / B);
for (j = 0; j < l; j++)
Rprintf("=");
Rprintf(">");
for (k = j; k < 50; k++)
Rprintf(" ");
Rprintf("]");
/* % completed */
Rprintf(" %3d%% completed", j * 2);
/* To delete the previous line */
Rprintf("\r");
/* Flush all char in buffer */
/* fflush(stdout); */
}
}
if (get_trace(controls))
Rprintf("\n");
PutRNGstate();
Free(prob); Free(iweights); Free(iweightstmp);
UNPROTECT(6);
return(ans);
}
/**
An experimental implementation of random forest like algorithms \n
*\param learnsample an object of class `LearningSample'
*\param weights a vector of case weights
*\param bwhere integer matrix (n x ntree) for terminal node numbers
*\param bweights double matrix (n x ntree) for bootstrap case weights
*\param fitmem an object of class `TreeFitMemory'
*\param controls an object of class `TreeControl'
*/
SEXP R_Ensemble_weights(SEXP learnsample, SEXP bweights,
SEXP controls) {
SEXP nweights, tree, where, ensemble, fitmem, ans, bwhere;
double *dnweights, *dweights;
int nobs, i, b, B , nodenum = 1, *iwhere;
int j, k, l;
B = get_ntree(controls);
nobs = get_nobs(learnsample);
PROTECT(ans = party_NEW_OBJECT("RandomForest"));
PROTECT(bwhere = allocVector(VECSXP, B));
PROTECT(ensemble = allocVector(VECSXP, B));
PROTECT(fitmem = ctree_memory(learnsample, PROTECT(ScalarLogical(1))));
SET_SLOT(ans, PL2_ensembleSym, ensemble);
SET_SLOT(ans, PL2_whereSym, bwhere);
SET_SLOT(ans, PL2_weightsSym, bweights);
/* <FIXME> can we call those guys ONCE? what about the deeper
calls??? </FIXME> */
GetRNGstate();
if (get_trace(controls))
Rprintf("\n");
for (b = 0; b < B; b++) {
SET_VECTOR_ELT(ensemble, b, tree = allocVector(VECSXP, NODE_LENGTH + 1));
SET_VECTOR_ELT(bwhere, b, where = allocVector(INTSXP, nobs));
iwhere = INTEGER(where);
for (i = 0; i < nobs; i++) iwhere[i] = 0;
C_init_node(tree, nobs, get_ninputs(learnsample),
get_maxsurrogate(get_splitctrl(controls)),
ncol(get_predict_trafo(GET_SLOT(learnsample,
PL2_responsesSym))));
nweights = S3get_nodeweights(tree);
dnweights = REAL(nweights);
dweights = REAL(VECTOR_ELT(bweights, b));
for (i = 0; i < nobs; i++) {
dnweights[i] = dweights[i];
}
C_TreeGrow(tree, learnsample, fitmem, controls, iwhere, &nodenum, 1);
nodenum = 1;
C_remove_weights(tree, 0);
/* compute terminal node ids for all observations (not just those with weight > 0) */
for (i = 0; i < nobs; i++)
iwhere[i] = C_get_nodeID(tree, GET_SLOT(learnsample, PL2_inputsSym), 0.0, i, -1);
if (get_trace(controls)) {
/* progress bar; inspired by
http://avinashjoshi.co.in/2009/10/13/creating-a-progress-bar-in-c/ */
Rprintf("[");
/* Print the = until the current percentage */
l = (int) ceil( ((double) b * 50.0) / B);
for (j = 0; j < l; j++)
Rprintf("=");
Rprintf(">");
for (k = j; k < 50; k++)
Rprintf(" ");
Rprintf("]");
/* % completed */
Rprintf(" %3d%% completed", j * 2);
/* To delete the previous line */
Rprintf("\r");
/* Flush all char in buffer */
/* fflush(stdout); */
}
}
if (get_trace(controls))
Rprintf("\n");
PutRNGstate();
UNPROTECT(5);
return(ans);
}
|