sam
2 years ago
parent
f434afbee6
commit
52ffb1ad25
Binary file not shown.
@ -0,0 +1,57 @@
|
||||
#ifndef CART_H
|
||||
#define CART_H
|
||||
|
||||
#include "Evaluation.h"
|
||||
|
||||
struct minEval;
|
||||
|
||||
struct DR;
|
||||
|
||||
struct DT;
|
||||
|
||||
//enum Evaluation {gini, entropy, logLoss};
|
||||
|
||||
class DecisionTree{
|
||||
public:
|
||||
|
||||
DT* DTree = nullptr;
|
||||
int maxHeight;
|
||||
long feature;
|
||||
long maxFeature;
|
||||
long seed;
|
||||
long classes;
|
||||
int* Sparse;
|
||||
double forgetRate;
|
||||
Evaluation evalue;
|
||||
long Rebuild;
|
||||
long roundNo;
|
||||
long called;
|
||||
long retain;
|
||||
|
||||
DecisionTree(int hight, long f, int* sparse, double forget, long maxFeature, long noClasses, Evaluation e, long r, long rb);
|
||||
|
||||
void Stablelize();
|
||||
|
||||
void Free();
|
||||
|
||||
minEval findMinGiniDense(double** data, long* result, long* totalT, long size, long col);
|
||||
|
||||
minEval findMinGiniSparse(double** data, long* result, long* totalT, long size, long col, DT* current);
|
||||
|
||||
minEval incrementalMinGiniDense(double** data, long* result, long size, long col, long*** count, double** record, long* max, long newCount, long forgetSize, bool isRoot);
|
||||
|
||||
minEval incrementalMinGiniSparse(double** dataNew, long* resultNew, long sizeNew, long sizeOld, DT* current, long col, long forgetSize, bool isRoot);
|
||||
|
||||
long* fitThenPredict(double** trainData, long* trainResult, long trainSize, double** testData, long testSize);
|
||||
|
||||
void fit(double** data, long* result, long size);
|
||||
|
||||
void Update(double** data, long* result, long size, DT* current);
|
||||
|
||||
void IncrementalUpdate(double** data, long* result, long size, DT* current);
|
||||
|
||||
long Test(double* data, DT* root);
|
||||
|
||||
void print(DT* root);
|
||||
};
|
||||
#endif
|
||||
@ -0,0 +1,278 @@
|
||||
#include "Evaluation.h"
|
||||
#include <cfloat>
|
||||
#include <math.h>
|
||||
#include <cstdio>
|
||||
|
||||
struct minEval{
|
||||
double value;
|
||||
double values;
|
||||
|
||||
double eval;
|
||||
long left; // how many on its left
|
||||
double* record;
|
||||
long max;
|
||||
long** count;
|
||||
long* sorted; // sorted d
|
||||
};
|
||||
|
||||
minEval giniSparse(double** data, long* result, long* d, long size, long col, long classes, long* totalT){
|
||||
double max = data[d[size-1]][col];
|
||||
minEval ret;
|
||||
ret.eval = DBL_MAX;
|
||||
|
||||
long i, j;
|
||||
long count[classes];
|
||||
long total = 0;
|
||||
for(i=0; i<classes; i++)count[i]=0;
|
||||
double gini1, gini2;
|
||||
double c;
|
||||
long l, r;
|
||||
for(i=0; i<size; i++){
|
||||
c = data[d[i]][col];
|
||||
if(c==max)break;
|
||||
count[result[d[i]]]++;
|
||||
total++;
|
||||
if(c==data[d[i+1]][col])continue;
|
||||
gini1 = 1.0;
|
||||
gini2 = 1.0;
|
||||
for(j=0;j<classes;j++){
|
||||
l = count[j];
|
||||
r = totalT[j]-l;
|
||||
gini1 -= pow((double)l/total, 2);
|
||||
gini2 -= pow((double)r/(size-total), 2);
|
||||
}
|
||||
gini1 = gini1*total/size + gini2*(size-total)/size;
|
||||
if(ret.eval>gini1){
|
||||
ret.eval = gini1;
|
||||
ret.value = c;
|
||||
ret.left = total;
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval entropySparse(double** data, long* result, long* d, long size, long col, long classes, long* totalT){
|
||||
double max = data[d[size-1]][col];
|
||||
minEval ret;
|
||||
ret.eval = DBL_MAX;
|
||||
|
||||
long i, j;
|
||||
long count[classes];
|
||||
long total = 0;
|
||||
for(i=0; i<classes; i++)count[i]=0;
|
||||
double entropy1, entropy2;
|
||||
double c;
|
||||
long l, r;
|
||||
for(i=0; i<size; i++){
|
||||
c = data[d[i]][col];
|
||||
if(c==max)break;
|
||||
count[result[d[i]]]++;
|
||||
total++;
|
||||
if(c==data[d[i+1]][col])continue;
|
||||
entropy1 = 0;
|
||||
entropy2 = 0;
|
||||
for(j=0;j<classes;j++){
|
||||
l = count[j];
|
||||
r = totalT[j]-l;
|
||||
entropy1 -= ((double)l/total)*log((double)l/total);
|
||||
entropy2 -= ((double)r/(size-total))*log((double)r/(size-total));
|
||||
}
|
||||
entropy1 = entropy1*total/size + entropy2*(size-total)/size;
|
||||
if(ret.eval>entropy1){
|
||||
ret.eval = entropy1;
|
||||
ret.value = c;
|
||||
ret.left = total;
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval giniSparseIncremental(long sizeTotal, long classes, double* newSortedData, long* newSortedResult, long* T){
|
||||
long l, r, i, j;
|
||||
minEval ret;
|
||||
ret.eval = DBL_MAX;
|
||||
double gini1, gini2;
|
||||
long count[classes];
|
||||
long total = 0;
|
||||
for(i=0; i<classes; i++)count[i]=0;
|
||||
double c, max=newSortedData[sizeTotal-1]; // largest value
|
||||
|
||||
for(i=0; i<sizeTotal; i++){
|
||||
c = newSortedData[i];
|
||||
if(c==max)break;
|
||||
count[newSortedResult[i]]++;
|
||||
total++;
|
||||
if(c==newSortedData[i+1])continue;
|
||||
gini1 = 1.0;
|
||||
gini2 = 1.0;
|
||||
for(j=0;j<classes;j++){
|
||||
l = count[j];
|
||||
r = T[j]-l;
|
||||
gini1 -= pow((double)l/total, 2);
|
||||
gini2 -= pow((double)r/(sizeTotal-total), 2);
|
||||
}
|
||||
gini1 = (gini1*total)/sizeTotal + (gini2*(sizeTotal-total))/sizeTotal;
|
||||
if(ret.eval>gini1){
|
||||
ret.eval = gini1;
|
||||
ret.value = c;
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval entropySparseIncremental(long sizeTotal, long classes, double* newSortedData, long* newSortedResult, long* T){
|
||||
long l, r, i, j;
|
||||
minEval ret;
|
||||
ret.eval = DBL_MAX;
|
||||
double e1, e2;
|
||||
long count[classes];
|
||||
long total = 0;
|
||||
for(i=0; i<classes; i++)count[i]=0;
|
||||
double c, max=newSortedData[sizeTotal-1]; // largest value
|
||||
|
||||
for(i=0; i<sizeTotal; i++){
|
||||
c = newSortedData[i];
|
||||
if(c==max)break;
|
||||
count[newSortedResult[i]]++;
|
||||
total++;
|
||||
if(c==newSortedData[i+1])continue;
|
||||
e1 = 0;
|
||||
e2 = 0;
|
||||
for(j=0;j<classes;j++){
|
||||
l = count[j];
|
||||
r = T[j]-l;
|
||||
e1 -= ((double)l/total)*log((double)l/total);
|
||||
e2 -= ((double)r/(sizeTotal-total))*log((double)r/(sizeTotal-total));
|
||||
}
|
||||
e1 = e1*total/sizeTotal + e2*(sizeTotal-total)/sizeTotal;
|
||||
if(ret.eval>e1){
|
||||
ret.eval = e1;
|
||||
ret.value = c;
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval giniDense(long max, long size, long classes, long** rem, long* d, double* record, long* totalT){
|
||||
minEval ret;
|
||||
ret.eval = DBL_MAX;
|
||||
|
||||
double gini1, gini2;
|
||||
long *t, *t2, *r, *r2, i, j;
|
||||
for(i=0;i<max;i++){
|
||||
t = rem[d[i]];
|
||||
if(i>0){
|
||||
t2 = rem[d[i-1]];
|
||||
for(j=0;j<=classes;j++){
|
||||
t[j]+=t2[j];
|
||||
}
|
||||
}
|
||||
if(t[classes]>=size)break;
|
||||
gini1 = 1.0;
|
||||
gini2 = 1.0;
|
||||
for(j=0;j<classes;j++){
|
||||
long l, r;
|
||||
l = t[j];
|
||||
r = totalT[j]-l;
|
||||
gini1 -= pow((double)l/t[classes], 2);
|
||||
gini2 -= pow((double)r/(size-t[classes]), 2);
|
||||
}
|
||||
gini1 = (gini1*t[classes])/size + (gini2*(size-t[classes]))/size;
|
||||
if(gini1<ret.eval){
|
||||
ret.eval = gini1;
|
||||
ret.value = record[d[i]];
|
||||
ret.left = t[classes];
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval entropyDense(long max, long size, long classes, long** rem, long* d, double* record, long* totalT){
|
||||
minEval ret;
|
||||
ret.eval = DBL_MAX;
|
||||
|
||||
double entropy1, entropy2;
|
||||
long *t, *t2, *r, *r2, i, j;
|
||||
for(i=0;i<max;i++){
|
||||
t = rem[d[i]];
|
||||
if(i>0){
|
||||
t2 = rem[d[i-1]];
|
||||
for(j=0;j<=classes;j++){
|
||||
t[j]+=t2[j];
|
||||
}
|
||||
}
|
||||
if(t[classes]>=size)break;
|
||||
entropy1 = 0;
|
||||
entropy2 = 0;
|
||||
for(j=0;j<classes;j++){
|
||||
long l, r;
|
||||
l = t[j];
|
||||
r = totalT[j]-l;
|
||||
entropy1 -= ((double)l/t[classes])*log((double)l/t[classes]);
|
||||
entropy2 -= ((double)r/(size-t[classes]))*log((double)r/(size-t[classes]));
|
||||
}
|
||||
entropy1 = entropy1*t[classes]/size + entropy2*(size-t[classes])/size;
|
||||
if(entropy1<ret.eval){
|
||||
ret.eval = entropy1;
|
||||
ret.value = record[d[i]];
|
||||
ret.left = t[classes];
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval giniDenseIncremental(long max, double* record, long** count, long classes, long newSize, long* T){
|
||||
double gini1, gini2;
|
||||
minEval ret;
|
||||
long i, j;
|
||||
|
||||
ret.eval = DBL_MAX;
|
||||
for(i=0; i<max; i++){
|
||||
if(count[i][classes]==newSize){
|
||||
continue;
|
||||
}
|
||||
gini1 = 1.0;
|
||||
gini2 = 1.0;
|
||||
for(j=0;j<classes;j++){
|
||||
long l, r;
|
||||
l = count[i][j];
|
||||
r = T[j]-l;
|
||||
gini1 -= pow((double)l/count[i][classes], 2);
|
||||
gini2 -= pow((double)r/(newSize-count[i][classes]), 2);
|
||||
}
|
||||
gini1 = gini1*count[i][classes]/newSize + gini2*((newSize-count[i][classes]))/newSize;
|
||||
if(gini1<ret.eval){
|
||||
ret.eval = gini1;
|
||||
ret.value = record[i];
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval entropyDenseIncremental(long max, double* record, long** count, long classes, long newSize, long* T){
|
||||
double entropy1, entropy2;
|
||||
minEval ret;
|
||||
long i, j;
|
||||
|
||||
ret.eval = DBL_MAX;
|
||||
for(i=0; i<max; i++){
|
||||
if(count[i][classes]==newSize or count[i][classes]==0){
|
||||
continue;
|
||||
}
|
||||
entropy1 = 0;
|
||||
entropy2 = 0;
|
||||
for(j=0;j<classes;j++){
|
||||
long l, r;
|
||||
l = count[i][j];
|
||||
r = T[j]-l;
|
||||
entropy1 -= ((double)l/count[i][classes])*log((double)l/count[i][classes]);
|
||||
entropy2 -= (double)r/(newSize-count[i][classes])*log((double)r/(newSize-count[i][classes]));
|
||||
}
|
||||
entropy1 = entropy1*count[i][classes]/newSize + entropy2*((newSize-count[i][classes]))/newSize;
|
||||
if(entropy1<ret.eval){
|
||||
ret.eval = entropy1;
|
||||
ret.value = record[i];
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
@ -0,0 +1,24 @@
|
||||
#ifndef EVALUATION_H
|
||||
#define EVALUATION_H
|
||||
|
||||
struct minEval;
|
||||
|
||||
enum Evaluation {gini, entropy, logLoss};
|
||||
|
||||
minEval giniSparse(double** data, long* result, long* d, long size, long col, long classes, long* totalT);
|
||||
|
||||
minEval entropySparse(double** data, long* result, long* d, long size, long col, long classes, long* totalT);
|
||||
|
||||
minEval giniSparseIncremental(long sizeTotal, long classes, double* newSortedData, long* newSortedResult, long* T);
|
||||
|
||||
minEval entropySparseIncremental(long sizeTotal, long classes, double* newSortedData, long* newSortedResult, long* T);
|
||||
|
||||
minEval giniDense(long max, long size, long classes, long** rem, long* d, double* record, long* totalT);
|
||||
|
||||
minEval entropyDense(long max, long size, long classes, long** rem, long* d, double* record, long* totalT);
|
||||
|
||||
minEval giniDenseIncremental(long max, double* record, long** count, long classes, long newSize, long* T);
|
||||
|
||||
minEval entropyDenseIncremental(long max, double* record, long** count, long classes, long newSize, long* T);
|
||||
|
||||
#endif
|
||||
@ -0,0 +1,139 @@
|
||||
#include "RF.h"
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
#include <ctime>
|
||||
|
||||
struct DT{
|
||||
int height;
|
||||
long* featureId;
|
||||
DT* left = nullptr;
|
||||
DT* right = nullptr;
|
||||
|
||||
// split info
|
||||
bool terminate;
|
||||
double dpoint;
|
||||
long feature;
|
||||
long result;
|
||||
|
||||
// Sparse data record
|
||||
double** sortedData; // for each feature, sorted data
|
||||
long** sortedResult;
|
||||
|
||||
// Dense data record
|
||||
long*** count = nullptr;// for each feature, number of data belongs to each class and dense value
|
||||
double** record = nullptr;// for each feature, record each dense data
|
||||
long* max = nullptr;// number of dense value of each feature
|
||||
|
||||
//long* T; // number of data in each class in this node
|
||||
double** dataRecord = nullptr;// Record the data
|
||||
long* resultRecord = nullptr;// Record the result
|
||||
long size = 0;// Size of the dataset
|
||||
};
|
||||
|
||||
RandomForest::RandomForest(long mTree, long actTree, long rTime, int h, long feature, int* s, double forg, long maxF, long noC, Evaluation eval, long r, long rb){
|
||||
srand((long)clock());
|
||||
Rebuild = rb;
|
||||
if(actTree<1)actTree=1;
|
||||
noTree = actTree;
|
||||
activeTree = actTree;
|
||||
treePointer = 0;
|
||||
if(mTree<actTree)mTree=activeTree;
|
||||
maxTree = mTree;
|
||||
if(rTime<=0)rTime=1;
|
||||
rotateTime = rTime;
|
||||
timer = 0;
|
||||
retain = r;
|
||||
|
||||
long i;
|
||||
height = h;
|
||||
f = feature;
|
||||
sparse = new int[f];
|
||||
for(i=0; i<f; i++)sparse[i]=s[i];
|
||||
forget = forg;
|
||||
maxFeature = maxF;
|
||||
noClasses = noC;
|
||||
e = eval;
|
||||
|
||||
DTrees = (DecisionTree**)malloc(mTree*sizeof(DecisionTree*));
|
||||
for(i=0; i<mTree; i++){
|
||||
if(i<actTree){
|
||||
DTrees[i] = new DecisionTree(height, f, sparse, forget, maxFeature, noClasses, e, r, rb);
|
||||
}
|
||||
else{
|
||||
DTrees[i]=nullptr;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void RandomForest::fit(double** data, long* result, long size){
|
||||
if(timer==rotateTime and maxTree!=activeTree){
|
||||
Rotate();
|
||||
timer=0;
|
||||
}
|
||||
long i, j, k;
|
||||
double** newData;
|
||||
long* newResult;
|
||||
for(i=0; i<activeTree; i++){
|
||||
newData = new double*[size];
|
||||
newResult = new long[size];
|
||||
for(j = 0; j<size; j++){
|
||||
newData[j] = new double[f];
|
||||
for(k=0; k<f; k++){
|
||||
newData[j][k] = data[j][k];
|
||||
}
|
||||
newResult[j] = result[j];
|
||||
}
|
||||
DTrees[(i+treePointer)%maxTree]->fit(newData, newResult, size);
|
||||
}
|
||||
timer++;
|
||||
}
|
||||
|
||||
long* RandomForest::fitThenPredict(double** trainData, long* trainResult, long trainSize, double** testData, long testSize){
|
||||
fit(trainData, trainResult, trainSize);
|
||||
long* testResult = (long*)malloc(testSize*sizeof(long));
|
||||
for(long i=0; i<testSize; i++){
|
||||
testResult[i] = Test(testData[i]);
|
||||
}
|
||||
return testResult;
|
||||
}
|
||||
|
||||
void RandomForest::Rotate(){
|
||||
if(noTree==maxTree){
|
||||
DTrees[(treePointer+activeTree)%maxTree]->Free();
|
||||
delete DTrees[(treePointer+activeTree)%maxTree];
|
||||
}else{
|
||||
noTree++;
|
||||
}
|
||||
DTrees[(treePointer+activeTree)%maxTree] = new DecisionTree(height, f, sparse, forget, maxFeature, noClasses, e, retain, Rebuild);
|
||||
long size = DTrees[(treePointer+activeTree-1)%maxTree]->DTree->size;
|
||||
double** newData = new double*[size];
|
||||
long* newResult = new long[size];
|
||||
for(long j = 0; j<size; j++){
|
||||
newData[j] = new double[f];
|
||||
for(long k=0; k<f; k++){
|
||||
newData[j][k] = DTrees[(treePointer+activeTree-1)%maxTree]->DTree->dataRecord[j][k];
|
||||
}
|
||||
newResult[j] = DTrees[(treePointer+activeTree-1)%maxTree]->DTree->resultRecord[j];
|
||||
}
|
||||
|
||||
DTrees[(treePointer+activeTree)%maxTree]->fit(newData, newResult, size);
|
||||
DTrees[treePointer]->Stablelize();
|
||||
if(++treePointer==maxTree)treePointer=0;
|
||||
}
|
||||
|
||||
|
||||
long RandomForest::Test(double* data){
|
||||
long i;
|
||||
long predict[noClasses];
|
||||
for(i=0; i<noClasses; i++)predict[i]=0;
|
||||
for(i=0; i<noTree; i++){
|
||||
predict[DTrees[i]->Test(data, DTrees[i]->DTree)]++;
|
||||
}
|
||||
|
||||
long ret = 0;
|
||||
for(i=1; i<noClasses; i++){
|
||||
if(predict[i]>predict[ret])ret = i;
|
||||
}
|
||||
|
||||
return ret;
|
||||
}
|
||||
@ -0,0 +1,46 @@
|
||||
#ifndef RF_H
|
||||
#define RF_H
|
||||
|
||||
#include "DecisionTree.h"
|
||||
|
||||
struct minEval;
|
||||
|
||||
struct DR;
|
||||
|
||||
struct DT;
|
||||
|
||||
//enum Evaluation {gini, entropy, logLoss};
|
||||
|
||||
class RandomForest{
|
||||
public:
|
||||
|
||||
long noTree;
|
||||
long maxTree;
|
||||
long activeTree;
|
||||
long treePointer;
|
||||
long rotateTime;
|
||||
long timer;
|
||||
long retain;
|
||||
DecisionTree** DTrees = nullptr;
|
||||
|
||||
long height;
|
||||
long Rebuild;
|
||||
long f;
|
||||
int* sparse;
|
||||
double forget;
|
||||
long maxFeature;
|
||||
long noClasses;
|
||||
Evaluation e;
|
||||
|
||||
|
||||
RandomForest(long maxTree, long activeTree, long rotateTime, int height, long f, int* sparse, double forget, long maxFeature=0, long noClasses=2, Evaluation e=Evaluation::gini, long r=-1, long rb=2147483647);
|
||||
|
||||
void fit(double** data, long* result, long size);
|
||||
|
||||
long* fitThenPredict(double** trainData, long* trainResult, long trainSize, double** testData, long testSize);
|
||||
|
||||
void Rotate();
|
||||
|
||||
long Test(double* data);
|
||||
};
|
||||
#endif
|
||||
@ -0,0 +1,774 @@
|
||||
#include <algorithm>
|
||||
#include <stddef.h>
|
||||
#include <stdlib.h>
|
||||
#include <math.h>
|
||||
#include "DecisionTree.h"
|
||||
#include "Evaluation.h"
|
||||
#include <fstream>
|
||||
#include <float.h>
|
||||
#include <ctime>
|
||||
|
||||
struct minEval{
|
||||
double value;
|
||||
int* values;
|
||||
|
||||
double eval;
|
||||
long left; // how many on its left
|
||||
double* record;
|
||||
long max;
|
||||
long** count;
|
||||
};
|
||||
|
||||
struct DT{
|
||||
int height;
|
||||
long* featureId;
|
||||
DT* left = nullptr;
|
||||
DT* right = nullptr;
|
||||
|
||||
// split info
|
||||
bool terminate;
|
||||
double dpoint;
|
||||
long feature;
|
||||
long result;
|
||||
|
||||
// Sparse data record
|
||||
double** sortedData; // for each feature, sorted data
|
||||
long** sortedResult;
|
||||
|
||||
// Dense data record
|
||||
long*** count = nullptr;// for each feature, number of data belongs to each class and dense value
|
||||
double** record = nullptr;// for each feature, record each dense data
|
||||
long* max = nullptr;// number of dense value of each feature
|
||||
|
||||
//long* T; // number of data in each class in this node
|
||||
double** dataRecord = nullptr;// Record the data
|
||||
long* resultRecord = nullptr;// Record the result
|
||||
long size = 0;// Size of the dataset
|
||||
};
|
||||
long seed = (long)clock();
|
||||
long* Rands(long feature, long maxFeature){
|
||||
//srand(seed++);
|
||||
long i;
|
||||
long* ret = (long*) malloc(feature*sizeof(long));
|
||||
for(i =0; i<feature; i++)ret[i] = i;
|
||||
if(maxFeature==feature){
|
||||
return ret;
|
||||
}
|
||||
std::random_shuffle(ret, &ret[feature]);
|
||||
long* ret2 = (long*) malloc(maxFeature*sizeof(long));
|
||||
for(i=0; i<maxFeature; i++)ret2[i] = ret[i];
|
||||
free(ret);
|
||||
return ret2;
|
||||
}
|
||||
double getRand(){
|
||||
return (double) rand() / RAND_MAX;
|
||||
}
|
||||
|
||||
|
||||
void createTree(DT* t, long currentHeight, long height, long f, long maxF, long classes){
|
||||
srand(seed);
|
||||
long i;
|
||||
t->count = (long***)malloc(f*sizeof(long**));
|
||||
for(i=0; i<f; i++)t->count[i]=nullptr;
|
||||
t->record = (double**)malloc(f*sizeof(double*));
|
||||
for(i=0; i<f; i++)t->record[i]=nullptr;
|
||||
t->max = (long*)malloc(f*sizeof(long));
|
||||
t->max[0] = -1;
|
||||
t->featureId = Rands(f, maxF);
|
||||
//t->T = (long*)malloc(classes*sizeof(long));
|
||||
t->sortedData = (double**) malloc(f*sizeof(double*));
|
||||
for(i=0; i<f; i++)t->sortedData[i]=nullptr;
|
||||
t->sortedResult = (long**) malloc(f*sizeof(long*));
|
||||
for(i=0; i<f; i++)t->sortedResult[i]=nullptr;
|
||||
t->dataRecord = nullptr;
|
||||
t->resultRecord = nullptr;
|
||||
t->height = currentHeight;
|
||||
t->feature = -1;
|
||||
t->size = 0;
|
||||
if(currentHeight>height){
|
||||
t->right = nullptr;
|
||||
t->left = nullptr;
|
||||
return;
|
||||
}
|
||||
|
||||
t->left = (DT*)malloc(sizeof(DT));
|
||||
t->right = (DT*)malloc(sizeof(DT));
|
||||
createTree(t->left, currentHeight+1, height, f, maxF, classes);
|
||||
createTree(t->right, currentHeight+1, height, f, maxF, classes);
|
||||
}
|
||||
|
||||
void stableTree(DT* t, long f){
|
||||
long i, j;
|
||||
for(i=0; i<f; i++){
|
||||
if(t->count[i]==nullptr)continue;
|
||||
for(j=0; j<t->max[i]; j++){
|
||||
free(t->count[i][j]);
|
||||
}
|
||||
free(t->count[i]);
|
||||
}
|
||||
free(t->count);
|
||||
for(i=0; i<f; i++){
|
||||
if(t->record[i]==nullptr)continue;
|
||||
free(t->record[i]);
|
||||
}
|
||||
free(t->record);
|
||||
free(t->max);
|
||||
free(t->featureId);
|
||||
for(i=0; i<f; i++){
|
||||
if(t->sortedData[i]==nullptr)continue;
|
||||
free(t->sortedData[i]);
|
||||
}
|
||||
free(t->sortedData);
|
||||
for(i=0; i<f; i++){
|
||||
if(t->sortedResult[i]==nullptr)continue;
|
||||
free(t->sortedResult[i]);
|
||||
}
|
||||
free(t->sortedResult);
|
||||
free(t->dataRecord);
|
||||
free(t->resultRecord);
|
||||
if(t->right!=nullptr)stableTree(t->right, f);
|
||||
if(t->left!=nullptr)stableTree(t->left, f);
|
||||
}
|
||||
|
||||
void freeTree(DT* t){
|
||||
if(t->left != nullptr)freeTree(t->left);
|
||||
if(t->right != nullptr)freeTree(t->right);
|
||||
free(t);
|
||||
}
|
||||
|
||||
DecisionTree::DecisionTree(int height, long f, int* sparse, double forget=0.1, long maxF=0, long noClasses=2, Evaluation e=Evaluation::gini, long r=-1, long rb=1){
|
||||
evalue = e;
|
||||
called = 0;
|
||||
long i;
|
||||
// Max tree height
|
||||
maxHeight = height;
|
||||
// Number of features
|
||||
feature = f;
|
||||
// If each feature is sparse or dense, 0 for dense, 1 for sparse, >2 for number of category
|
||||
Sparse = (int*)malloc(f*sizeof(int));
|
||||
for(i = 0; i<f; i++){
|
||||
Sparse[i] = sparse[i];
|
||||
}
|
||||
// Create Decision tree
|
||||
DTree = (DT*)malloc(sizeof(DT));
|
||||
DTree->feature = -1;
|
||||
// The number of feature that is considered in each node
|
||||
if(maxF>=f){
|
||||
maxFeature = f;
|
||||
}else if(maxF<=0){
|
||||
maxFeature = (long)round(sqrt(f));
|
||||
}else{
|
||||
maxFeature = maxF;
|
||||
}
|
||||
forgetRate = std::min(1.0, forget);
|
||||
retain = r;
|
||||
createTree(DTree, 0, maxHeight, f, maxFeature, noClasses);
|
||||
// Randomly generate the features
|
||||
//DTree->featureId = Rands();
|
||||
//DTree->sorted = (long**) malloc(f*sizeof(long*));
|
||||
// Number of classes of this dataset
|
||||
Rebuild = rb;
|
||||
roundNo = 0;
|
||||
classes = std::max(noClasses, (long)2);
|
||||
//DTree->T = (long*) malloc(noClasses*sizeof(long));
|
||||
/*for(long i = 0; i<noClasses; i++){
|
||||
DTree->T[i]=0;
|
||||
}*/
|
||||
}
|
||||
|
||||
void DecisionTree::Stablelize(){
|
||||
free(Sparse);
|
||||
stableTree(DTree, feature);
|
||||
}
|
||||
|
||||
void DecisionTree::Free(){
|
||||
freeTree(DTree);
|
||||
}
|
||||
|
||||
minEval DecisionTree::incrementalMinGiniSparse(double** dataTotal, long* resultTotal, long sizeTotal, long sizeNew, DT* current, long col, long forgetSize, bool isRoot){
|
||||
long i, j;
|
||||
if(isRoot){sizeNew=sizeTotal-forgetSize;}
|
||||
long newD[sizeNew];
|
||||
for(i=0; i<sizeNew; i++)newD[i]=i;
|
||||
long T[classes];
|
||||
for(i=0; i<classes; i++)T[i]=0;
|
||||
std::sort(newD, newD+sizeNew, [&dataTotal, col](long l, long r){return dataTotal[l][col]<dataTotal[r][col];});
|
||||
double* newSortedData = (double*)malloc(sizeTotal*sizeof(double));
|
||||
long* newSortedResult = (long*)malloc(sizeTotal*sizeof(long));
|
||||
long p1=0, p2=0;
|
||||
double* oldData = current->sortedData[col];
|
||||
long* oldResult = current->sortedResult[col];
|
||||
for(i=0; i<sizeTotal; i++){
|
||||
if(p1==sizeNew){
|
||||
newSortedData[i] = oldData[p2];
|
||||
newSortedResult[i] = oldResult[p2];
|
||||
T[newSortedResult[i]]++;
|
||||
p2++;
|
||||
}
|
||||
else if(p2==sizeTotal-sizeNew){
|
||||
newSortedData[i] = dataTotal[newD[p1]][col];
|
||||
newSortedResult[i] = resultTotal[newD[p1]];
|
||||
T[newSortedResult[i]]++;
|
||||
p1++;
|
||||
}
|
||||
else if(dataTotal[newD[p1]][col]<oldData[p2]){
|
||||
newSortedData[i] = dataTotal[newD[p1]][col];
|
||||
newSortedResult[i] = resultTotal[newD[p1]];
|
||||
T[newSortedResult[i]]++;
|
||||
p1++;
|
||||
}else{
|
||||
newSortedData[i] = oldData[p2];
|
||||
newSortedResult[i] = oldResult[p2];
|
||||
T[newSortedResult[i]]++;
|
||||
p2++;
|
||||
}
|
||||
}
|
||||
current->sortedData[col] = newSortedData;
|
||||
current->sortedResult[col] = newSortedResult;
|
||||
free(oldData);
|
||||
free(oldResult);
|
||||
|
||||
minEval ret;
|
||||
if(evalue == Evaluation::gini){
|
||||
ret = giniSparseIncremental(sizeTotal, classes, newSortedData, newSortedResult, T);
|
||||
}else if(evalue == Evaluation::entropy or evalue == Evaluation::logLoss){
|
||||
ret = entropySparseIncremental(sizeTotal, classes, newSortedData, newSortedResult, T);
|
||||
}
|
||||
ret.values = nullptr;
|
||||
return ret;
|
||||
}
|
||||
minEval DecisionTree::incrementalMinGiniDense(double** data, long* result, long size, long col, long*** count, double** record, long* max, long newSize, long forgetSize, bool isRoot){
|
||||
// newSize is before forget
|
||||
long low = 0;
|
||||
if(isRoot)size=newSize-forgetSize;
|
||||
long i, j, k;
|
||||
long newMax = 0;
|
||||
long maxLocal = max[col];
|
||||
long **newCount=(long**)malloc(size*sizeof(long*));
|
||||
for(i=0;i<size;i++){
|
||||
newCount[i] = (long*)malloc((classes+1)*sizeof(long));
|
||||
for(j=0;j<= classes;j++)newCount[i][j]=0;
|
||||
}
|
||||
double newRecord[size];
|
||||
bool find;
|
||||
|
||||
// find total count for each class
|
||||
long T[classes];
|
||||
for(i=0;i<classes;i++)T[i]=0;
|
||||
for(i=0;i<max[col];i++){
|
||||
for(j=0;j<classes;j++){
|
||||
if(isRoot)count[col][i][j]=0;
|
||||
else if(T[j]<count[col][i][j])T[j]=count[col][i][j];
|
||||
}
|
||||
}
|
||||
|
||||
// plug in new data
|
||||
for(i=0; i<size; i++){
|
||||
find = false;
|
||||
T[result[i]]++;
|
||||
for(j=0;j<max[col];j++){
|
||||
if(data[i][col]==record[col][j]){
|
||||
count[col][j][result[i]]++;
|
||||
count[col][j][classes] ++;
|
||||
find = true;
|
||||
}else if(data[i][col]<record[col][j]){
|
||||
count[col][j][result[i]]++;
|
||||
count[col][j][classes] ++;
|
||||
}
|
||||
}
|
||||
for(j=0;j<newMax;j++){
|
||||
if(data[i][col]==newRecord[j]){
|
||||
newCount[j][result[i]]++;
|
||||
newCount[j][classes] ++;
|
||||
find = true;
|
||||
} else if(data[i][col]<newRecord[j]){
|
||||
newCount[j][result[i]]++;
|
||||
newCount[j][classes] ++;
|
||||
}
|
||||
}
|
||||
if(not find){
|
||||
newRecord[newMax] = data[i][col];
|
||||
double currentMinMax = -1*DBL_MAX;
|
||||
for(j=0;j<max[col];j++){
|
||||
if(record[col][j]<newRecord[newMax] and record[col][j]>currentMinMax){
|
||||
currentMinMax = record[col][j];
|
||||
for(k=0;k<=classes;k++)newCount[newMax][k]=count[col][j][k];
|
||||
}
|
||||
}
|
||||
for(j=0;j<newMax;j++){
|
||||
if(newRecord[j]<newRecord[newMax] and currentMinMax<newRecord[j]){
|
||||
currentMinMax = newRecord[j];
|
||||
for(k=0;k<=classes;k++)newCount[newMax][k]=newCount[j][k];
|
||||
}
|
||||
}
|
||||
if(currentMinMax== -1*DBL_MAX){
|
||||
for(k=0;k<=classes;k++)newCount[newMax][k]=0;
|
||||
}
|
||||
newCount[newMax][result[i]]++;
|
||||
newCount[newMax][classes]++;
|
||||
newMax++;
|
||||
}
|
||||
}
|
||||
// Updata new count and record
|
||||
if(newMax>0){
|
||||
max[col]+=newMax;
|
||||
long** updateCount = (long**)malloc(max[col]*sizeof(long*));
|
||||
double* updateRecord = (double*)malloc(max[col]*sizeof(double));
|
||||
for(i=0; i<max[col]; i++){
|
||||
if(i>=newMax){
|
||||
updateCount[i] = count[col][i-newMax];
|
||||
updateRecord[i] = record[col][i-newMax];
|
||||
}
|
||||
else{
|
||||
updateCount[i] = newCount[i];
|
||||
updateRecord[i] = newRecord[i];
|
||||
}
|
||||
}
|
||||
free(count[col]);
|
||||
free(record[col]);
|
||||
count[col]=updateCount;
|
||||
record[col]=updateRecord;
|
||||
}
|
||||
for(i=newMax; i<size; i++){
|
||||
free(newCount[i]);
|
||||
}
|
||||
free(newCount);
|
||||
|
||||
//calculate gini
|
||||
minEval ret;
|
||||
if(evalue==Evaluation::gini){
|
||||
ret = giniDenseIncremental(max[col], record[col], count[col], classes, newSize, T);
|
||||
}else if(evalue==Evaluation::entropy or evalue==Evaluation::logLoss){
|
||||
ret = entropyDenseIncremental(max[col], record[col], count[col], classes, newSize, T);
|
||||
}
|
||||
ret.values = nullptr;
|
||||
return ret;
|
||||
}
|
||||
|
||||
minEval DecisionTree::findMinGiniSparse(double** data, long* result, long* totalT, long size, long col, DT* current){
|
||||
long i, j;
|
||||
long* d = (long*)malloc(size*sizeof(long));
|
||||
for(i=0; i<size; i++)d[i]=i;
|
||||
std::sort(d, d+size, [&data, col](long l, long r){return data[l][col]<data[r][col];});
|
||||
|
||||
minEval ret;
|
||||
if(evalue == Evaluation::gini){
|
||||
ret = giniSparse(data, result, d, size, col, classes, totalT);
|
||||
}else if(evalue == Evaluation::entropy or evalue == Evaluation::logLoss){
|
||||
ret = entropySparse(data, result, d, size, col, classes, totalT);
|
||||
}
|
||||
if(current->sortedData[col] != nullptr)free(current->sortedData[col]);
|
||||
if(current->sortedResult[col] != nullptr)free(current->sortedResult[col]);
|
||||
current->sortedData[col] = (double*) malloc(size*sizeof(double));
|
||||
current->sortedResult[col] = (long*) malloc(size*sizeof(long));
|
||||
for(i=0;i<size; i++){
|
||||
current->sortedData[col][i] = data[d[i]][col];
|
||||
current->sortedResult[col][i] = result[d[i]];
|
||||
}
|
||||
free(d);
|
||||
ret.values = nullptr;
|
||||
return ret;
|
||||
|
||||
}
|
||||
|
||||
minEval DecisionTree::findMinGiniDense(double** data, long* result, long* totalT, long size, long col){
|
||||
long low = 0;
|
||||
long i, j, k, max=0;
|
||||
long** count = (long**)malloc(size*sizeof(long*));
|
||||
// size2 and count2 are after forget
|
||||
double* record = (double*)malloc(size*sizeof(double));
|
||||
bool find;
|
||||
for(i=0;i<size;i++){
|
||||
find = false;
|
||||
for(j=0; j<max; j++){
|
||||
if(record[j]==data[i][col]){
|
||||
count[j][result[i]]++;
|
||||
count[j][classes]++;
|
||||
find = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if(not find){
|
||||
count[max] = (long*)malloc((classes+1)*sizeof(long));
|
||||
record[max]=data[i][col];
|
||||
for(j=0;j<=classes;j++){
|
||||
count[max][j] = 0;
|
||||
}
|
||||
count[max][result[i]]++;
|
||||
count[max][classes]++;
|
||||
max++;
|
||||
}
|
||||
}
|
||||
long** rem = (long**)malloc(max*sizeof(long*));
|
||||
double* record2 = (double*)malloc(max*sizeof(double));
|
||||
for(i=0;i<max;i++){
|
||||
rem[i] = count[i];
|
||||
record2[i] = record[i];
|
||||
}
|
||||
free(count);
|
||||
free(record);
|
||||
|
||||
long d[max];
|
||||
for(i=0;i<max;i++){
|
||||
d[i] = i;
|
||||
}
|
||||
std::sort(d, d+max, [&record2](long l, long r){return record2[l]<record2[r];});
|
||||
minEval ret;
|
||||
if(evalue == Evaluation::gini){
|
||||
ret = giniDense(max, size, classes, rem, d, record2, totalT);
|
||||
}else if(evalue == Evaluation::entropy or evalue == Evaluation::logLoss){
|
||||
ret = entropyDense(max, size, classes, rem, d, record2, totalT);
|
||||
}
|
||||
ret.record = record2;
|
||||
ret.max = max;
|
||||
ret.count = rem;
|
||||
ret.values = nullptr;
|
||||
return ret;
|
||||
}
|
||||
|
||||
double xxx;
|
||||
void DecisionTree::fit(double** data, long* result, long size){
|
||||
roundNo++;
|
||||
if(DTree->size==0){
|
||||
Update(data, result, size, DTree);
|
||||
}else{
|
||||
IncrementalUpdate(data, result, size, DTree);
|
||||
}
|
||||
/*
|
||||
if(Rebuild and called==10){
|
||||
called = 0;
|
||||
Rebuild = false;
|
||||
}else if(Rebuild){
|
||||
called = 11;
|
||||
}else{
|
||||
called++;
|
||||
}*/
|
||||
}
|
||||
|
||||
long* DecisionTree::fitThenPredict(double** trainData, long* trainResult, long trainSize, double** testData, long testSize){
|
||||
fit(trainData, trainResult, trainSize);
|
||||
long* testResult = (long*)malloc(testSize*sizeof(long));
|
||||
for(long i=0; i<testSize; i++){
|
||||
testResult[i] = Test(testData[i], DTree);
|
||||
}
|
||||
return testResult;
|
||||
}
|
||||
|
||||
void DecisionTree::IncrementalUpdate(double** data, long* result, long size, DT* current){
|
||||
long i, j;
|
||||
long low = 0;
|
||||
long forgetSize=0;
|
||||
if(retain>0 and current->size+size>retain) forgetSize = std::min(current->size+size - retain, current->size);
|
||||
else if(retain<0) forgetSize = (long)current->size*forgetRate;
|
||||
long* index = new long[current->size];
|
||||
double** dataNew;
|
||||
long* resultNew;
|
||||
if(current->height == 0){
|
||||
dataNew = (double**)malloc((size+current->size-forgetSize)*sizeof(double*));
|
||||
resultNew = (long*)malloc((size+current->size-forgetSize)*sizeof(long));
|
||||
for(i=0;i<size;i++){
|
||||
dataNew[i] = data[i];
|
||||
resultNew[i] = result[i];
|
||||
}
|
||||
for(i=0; i<current->size; i++){
|
||||
index[i] = i;
|
||||
}
|
||||
std::random_shuffle(index, index+current->size);
|
||||
long x = 0;
|
||||
for(i=0;i<current->size;i++){
|
||||
if(i>=current->size-forgetSize){
|
||||
current->dataRecord[index[i]][feature-1] = DBL_MAX;
|
||||
|
||||
}else{
|
||||
dataNew[i+size] = current->dataRecord[index[i]];
|
||||
resultNew[i+size] = current->resultRecord[index[i]];
|
||||
}
|
||||
}
|
||||
}else{
|
||||
forgetSize = 0;
|
||||
dataNew = (double**)malloc((size+current->size)*sizeof(double*));
|
||||
resultNew = (long*)malloc((size+current->size)*sizeof(long));
|
||||
for(i=0;i<size;i++){
|
||||
dataNew[i] = data[i];
|
||||
resultNew[i] = result[i];
|
||||
}
|
||||
for(i=0;i<current->size;i++){
|
||||
if(current->dataRecord[i][feature-1] == DBL_MAX){
|
||||
forgetSize++;
|
||||
continue;
|
||||
}else{
|
||||
dataNew[i+size-forgetSize] = current->dataRecord[i];
|
||||
resultNew[i+size-forgetSize] = current->resultRecord[i];
|
||||
}
|
||||
}
|
||||
}
|
||||
free(data);
|
||||
free(result);
|
||||
current->size -= forgetSize;
|
||||
current->size += size;
|
||||
// end condition
|
||||
if(current->terminate or roundNo%Rebuild==0){
|
||||
if(current->height == 0){
|
||||
for(i=0; i<forgetSize; i++){
|
||||
free(current->dataRecord[index[current->size-size+i]]);
|
||||
}
|
||||
}
|
||||
delete(index);
|
||||
Update(dataNew, resultNew, current->size, current);
|
||||
return;
|
||||
}
|
||||
// find min gini
|
||||
minEval c, cMin;
|
||||
long cFeature;
|
||||
cMin.eval = DBL_MAX;
|
||||
cMin.values = nullptr;
|
||||
// TODO
|
||||
for(i=0;i<maxFeature; i++){
|
||||
if(Sparse[current->featureId[i]]==1){
|
||||
c = incrementalMinGiniSparse(dataNew, resultNew, current->size+forgetSize, size, current, current->featureId[i], forgetSize, false);
|
||||
}
|
||||
else if(Sparse[current->featureId[i]]==0){
|
||||
c = incrementalMinGiniDense(dataNew, resultNew, size, current->featureId[i], current->count, current->record, current->max, current->size+forgetSize, forgetSize, false);
|
||||
}else{
|
||||
//c = incrementalMinGiniCategorical();
|
||||
}
|
||||
if(c.eval<cMin.eval){
|
||||
cMin.eval = c.eval;
|
||||
cMin.value = c.value;
|
||||
if(cMin.values != nullptr)free(cMin.values);
|
||||
cMin.values = c.values;
|
||||
cFeature = current->featureId[i];
|
||||
}else if(c.values!=nullptr)free(c.values);
|
||||
}
|
||||
if(cMin.eval==DBL_MAX){
|
||||
current->terminate = true;
|
||||
long t[classes];
|
||||
for(i=0;i<classes;i++){
|
||||
t[i]=0;
|
||||
}
|
||||
for(i=low;i<low+size;i++){
|
||||
t[result[i]]++;
|
||||
}
|
||||
if(cMin.values!=nullptr)free(cMin.values);
|
||||
current->result = std::distance(t, std::max_element(t, t+classes));
|
||||
return;
|
||||
}
|
||||
//diverse data
|
||||
long ptL=0, ptR=0;
|
||||
double* t;
|
||||
long currentSize = current->size;
|
||||
//TODO:Discrete
|
||||
// Same diverse point as last time
|
||||
if(current->dpoint==cMin.value and current->feature==cFeature){
|
||||
long xxx = current->left->size;
|
||||
/*for(i=0; i<size; i++){
|
||||
if(dataNew[i][current->feature]<=current->dpoint){
|
||||
ptL++;
|
||||
}else{
|
||||
ptR++;
|
||||
}
|
||||
}*/
|
||||
ptL = size;
|
||||
ptR = size;
|
||||
long* resultL = (long*)malloc((ptL)*sizeof(long));
|
||||
long* resultR = (long*)malloc((ptR)*sizeof(long));
|
||||
double** dataL = (double**)malloc((ptL)*sizeof(double*));
|
||||
double** dataR = (double**)malloc((ptR)*sizeof(double*));
|
||||
ptL = 0;
|
||||
ptR = 0;
|
||||
for(i=0; i<size; i++){
|
||||
if(dataNew[i][current->feature]<=current->dpoint){
|
||||
dataL[ptL] = dataNew[i];
|
||||
resultL[ptL] = resultNew[i];
|
||||
ptL++;
|
||||
}else{
|
||||
dataR[ptR] = dataNew[i];
|
||||
resultR[ptR] = resultNew[i];
|
||||
ptR++;
|
||||
}
|
||||
}
|
||||
IncrementalUpdate(dataL, resultL, ptL, current->left);
|
||||
IncrementalUpdate(dataR, resultR, ptR, current->right);
|
||||
|
||||
if(current->height == 0){
|
||||
for(i=0; i<forgetSize; i++){
|
||||
free(current->dataRecord[index[current->size-size+i]]);
|
||||
}
|
||||
}
|
||||
delete(index);
|
||||
free(current->dataRecord);
|
||||
free(current->resultRecord);
|
||||
current->dataRecord = dataNew;
|
||||
current->resultRecord = resultNew;
|
||||
return;
|
||||
}
|
||||
|
||||
// Different diverse point
|
||||
current->feature = cFeature;
|
||||
current->dpoint = cMin.value;
|
||||
/*for(i=0; i<currentSize; i++){
|
||||
if(dataNew[i][current->feature]<=current->dpoint){
|
||||
ptL++;
|
||||
}else{
|
||||
ptR++;
|
||||
}
|
||||
}*/
|
||||
long* resultL = (long*)malloc(currentSize*sizeof(long));
|
||||
long* resultR = (long*)malloc(currentSize*sizeof(long));
|
||||
double** dataL = (double**)malloc(currentSize*sizeof(double*));
|
||||
double** dataR = (double**)malloc(currentSize*sizeof(double*));
|
||||
ptL = 0;
|
||||
ptR = 0;
|
||||
for(i=0; i<currentSize; i++){
|
||||
if(dataNew[i][current->feature]<=current->dpoint){
|
||||
dataL[ptL] = dataNew[i];
|
||||
resultL[ptL] = resultNew[i];
|
||||
ptL++;
|
||||
}else{
|
||||
dataR[ptR] = dataNew[i];
|
||||
resultR[ptR] = resultNew[i];
|
||||
ptR++;
|
||||
}
|
||||
}
|
||||
Update(dataL, resultL, ptL, current->left);
|
||||
Update(dataR, resultR, ptR, current->right);
|
||||
|
||||
if(current->height == 0){
|
||||
for(i=0; i<forgetSize; i++){
|
||||
free(current->dataRecord[index[current->size-size+i]]);
|
||||
}
|
||||
}
|
||||
|
||||
delete(index);
|
||||
free(current->dataRecord);
|
||||
free(current->resultRecord);
|
||||
current->dataRecord = dataNew;
|
||||
current->resultRecord = resultNew;
|
||||
}
|
||||
void DecisionTree::Update(double** data, long* result, long size, DT* current){
|
||||
long low = 0;
|
||||
long i, j;
|
||||
// end condition
|
||||
if(current->dataRecord!=nullptr)free(current->dataRecord);
|
||||
current->dataRecord = data;
|
||||
if(current->resultRecord!=nullptr)free(current->resultRecord);
|
||||
current->resultRecord = result;
|
||||
current->size = size;
|
||||
if(current->height == maxHeight){
|
||||
current->terminate = true;
|
||||
long t[classes];
|
||||
for(i=0;i<classes;i++){
|
||||
t[i]=0;
|
||||
}
|
||||
for(i=low;i<low+size;i++){
|
||||
t[result[i]]++;
|
||||
}
|
||||
current->result = std::distance(t, std::max_element(t, t+classes));
|
||||
return;
|
||||
}
|
||||
long T[classes];
|
||||
for(i=0;i<classes;i++){
|
||||
T[i] = 0;
|
||||
}
|
||||
for(i=0;i<size;i++){
|
||||
j = result[i];
|
||||
T[j]++;
|
||||
}
|
||||
for(i=0;i<classes;i++){
|
||||
if(T[i]==size){
|
||||
current->terminate = true;
|
||||
current->result = i;
|
||||
return;
|
||||
}
|
||||
}
|
||||
// find min Evaluation
|
||||
minEval c, cMin;
|
||||
long cFeature, oldMax, col, left=0;
|
||||
cMin.eval = DBL_MAX;
|
||||
cMin.values = nullptr;
|
||||
//TODO
|
||||
for(i=0;i<maxFeature; i++){
|
||||
col = current->featureId[i];
|
||||
if(Sparse[current->featureId[i]]==1){
|
||||
c = findMinGiniSparse(data, result, T, size, col, current);
|
||||
}
|
||||
else if(Sparse[current->featureId[i]]==0){
|
||||
c = findMinGiniDense(data, result, T, size, col);
|
||||
if(current->count[col]!=nullptr){
|
||||
for(j=0; j<current->max[col]; j++){
|
||||
if(current->count[col][j]!=nullptr)free(current->count[col][j]);
|
||||
}
|
||||
free(current->count[col]);
|
||||
free(current->record[col]);
|
||||
}
|
||||
current->count[col] = c.count;
|
||||
current->record[col] = c.record;
|
||||
current->max[col] = c.max;
|
||||
}else{
|
||||
|
||||
}
|
||||
if(c.eval<cMin.eval){
|
||||
cMin.eval = c.eval;
|
||||
if(cMin.values!=nullptr)free(cMin.values);
|
||||
cMin.values = c.values;
|
||||
cMin.value = c.value;
|
||||
cFeature = current->featureId[i];
|
||||
left = c.left;
|
||||
}else if(c.values!=nullptr){
|
||||
free(c.values);
|
||||
}
|
||||
}
|
||||
if(cMin.eval == DBL_MAX){
|
||||
current->terminate = true;
|
||||
long max = 0;
|
||||
for(i=1;i<classes;i++){
|
||||
if(T[max]<T[i])max=i;
|
||||
}
|
||||
if(cMin.values!=nullptr)free(cMin.values);
|
||||
current->result = max;
|
||||
return;
|
||||
}
|
||||
//diverse data
|
||||
current->terminate = false;
|
||||
current->feature = cFeature;
|
||||
current->dpoint = cMin.value;
|
||||
long ptL=0, ptR=0;
|
||||
//TODO:Discrete
|
||||
long* resultL = new long[left];
|
||||
long* resultR = new long[size-left];
|
||||
double** dataL = new double*[left];
|
||||
double** dataR = new double*[size-left];
|
||||
for(i=low; i<low+size; i++){
|
||||
if(data[i][current->feature]<=current->dpoint){
|
||||
dataL[ptL] = data[i];
|
||||
resultL[ptL] = result[i];
|
||||
ptL++;
|
||||
}else{
|
||||
dataR[ptR] = data[i];
|
||||
resultR[ptR] = result[i];
|
||||
ptR++;
|
||||
}
|
||||
}
|
||||
Update(dataL, resultL, ptL, current->left);
|
||||
Update(dataR, resultR, ptR, current->right);
|
||||
}
|
||||
|
||||
long DecisionTree::Test(double* data, DT* root){
|
||||
if(root->terminate)return root->result;
|
||||
if(data[root->feature]<=root->dpoint)return Test(data, root->left);
|
||||
return Test(data, root->right);
|
||||
}
|
||||
|
||||
void DecisionTree::print(DT* root){
|
||||
int x;
|
||||
//std::cin>>x;
|
||||
if(root->terminate){
|
||||
printf("%ld", root->result);
|
||||
return;
|
||||
}
|
||||
printf("([%ld, %f]:", root->feature, root->dpoint);
|
||||
print(root->left);
|
||||
printf(", ");
|
||||
print(root->right);
|
||||
printf(")");
|
||||
}
|
||||
@ -0,0 +1,53 @@
|
||||
#include "DecisionTree.h"
|
||||
#include "aquery.h"
|
||||
// __AQ_NO_SESSION__
|
||||
#include "../server/table.h"
|
||||
|
||||
DecisionTree* dt = nullptr;
|
||||
long pt = 0;
|
||||
double** data = nullptr;
|
||||
long* result = nullptr;
|
||||
|
||||
__AQEXPORT__(bool) newtree(int height, long f, ColRef<int> sparse, double forget, long maxf, long noclasses, Evaluation e, long r, long rb){
|
||||
if(sparse.size!=f)return 0;
|
||||
int* issparse = (int*)malloc(f*sizeof(int));
|
||||
for(long i=0; i<f; i++){
|
||||
issparse[i] = sparse.container[i];
|
||||
}
|
||||
if(maxf<0)maxf=f;
|
||||
dt = new DecisionTree(height, f, issparse, forget, maxf, noclasses, e, r, rb);
|
||||
return 1;
|
||||
}
|
||||
|
||||
__AQEXPORT__(bool) additem(ColRef<double>X, long y, long size){
|
||||
long j = 0;
|
||||
if(size>0){
|
||||
free(data);
|
||||
free(result);
|
||||
pt = 0;
|
||||
data=(double**)malloc(size*sizeof(double*));
|
||||
result=(long*)malloc(size*sizeof(long));
|
||||
}
|
||||
data[pt] = (double*)malloc(X.size*sizeof(double));
|
||||
for(j=0; j<X.size; j++){
|
||||
data[pt][j]=X.container[j];
|
||||
}
|
||||
result[pt] = y;
|
||||
return 1;
|
||||
}
|
||||
__AQEXPORT__(bool) fit(){
|
||||
if(pt<=0)return 0;
|
||||
dt->fit(data, result, pt);
|
||||
return 1;
|
||||
}
|
||||
|
||||
__AQEXPORT__(ColRef_storage) predict(){
|
||||
int* result = (int*)malloc(pt*sizeof(int));
|
||||
for(long i=0; i<pt; i++){
|
||||
result[i]=dt->Test(data[i], dt->DTree);
|
||||
}
|
||||
ColRef_storage ret(result, pt, pt, "prediction", 0);
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
||||
Binary file not shown.
@ -1,8 +1,22 @@
|
||||
LOAD MODULE FROM "./test.so"
|
||||
LOAD MODULE FROM "./libirf.so"
|
||||
FUNCTIONS (
|
||||
mydiv(a:int, b:int) -> double,
|
||||
mulvec(a:int, b:vecfloat) -> vecfloat
|
||||
newtree(height:int, f:int64, sparse:vecint, forget:double, maxf:int64, noclasses:int64, e:int, r:int64, rb:int64) -> bool,
|
||||
additem(X:vecdouble, y:int64, size:int64) -> bool,
|
||||
fit() -> bool,
|
||||
predict() -> vecint
|
||||
);
|
||||
|
||||
select mydiv(2,3);
|
||||
|
||||
create table tb(x int);
|
||||
create table tb2(x double, y double, z double);
|
||||
insert into tb values (0);
|
||||
insert into tb values (0);
|
||||
insert into tb values (0);
|
||||
select newtree(5, 3, tb.x, 0, 3, 2, 0, 100, 1) from tb;
|
||||
insert into tb2 values (1, 0, 1);
|
||||
insert into tb2 values (0, 1, 1);
|
||||
insert into tb2 values (1, 1, 1);
|
||||
select additem(tb2.x, 1, 3) from tb2;
|
||||
select additem(tb2.y, 0, -1) from tb2;
|
||||
select additem(tb2.z, 1, -1) from tb2;
|
||||
select fit();
|
||||
select predict();
|
||||
|
||||
Loading…
Reference in new issue