// matlab entry point
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) {
// Currently, we do not have any input or output
if (nrhs != 0)
mexErrMsgTxt("Wrong number of inputs");
if (nlhs != 0)
mexErrMsgTxt("Wrong number of outputs");
// set the seed for random numbers
srand(time(NULL));
// set parameters for learning
PARAMETER param;
SetParameters(param);
// get image feature descriptors from the file
vector<vector<SINGLEIM>> set_imVec;
int classNum;
ifstream fi("view_list.txt");
char ind[80];
while(fi >> ind )
{
char name[80] = "trainval_0_";
strcat(name, ind);
strcat(name,".txt");
mexPrintf("%s", name);
vector<SINGLEIM> imVec;
GetData(imVec, name, param);
set_imVec.push_back(imVec);
}
// GetData(imVec, "trainval_0_flipped.txt", param);
AdjustLabels(set_imVec, classNum);
mexPrintf("Finish getting data from the disk.\n");
// pre-pooling for the background information, if it is necessary
// if (param.bgPrePooling == 1)
// BgPrePooling(imVec, param);
// mexPrintf("Finish getting background feature.\n");
// the pre-allocated memory for all the trees
int treeMemSize = int(pow(2, param.maxTreeDepth + 1)) * param.treeNum;
TREENODE *treeMemory = new TREENODE[treeMemSize];
InitTreeMem(treeMemory, treeMemSize);
int treeMemID = 0;
vector<TREENODE*> trees; // the set of decision trees
for (int i = 0; i < param.treeNum; i++)
{
trees.push_back(treeMemory + treeMemID);
treeMemID++;
}
mexPrintf("Finish initializing all the trees.\n");
// train decision trees
clock_t t1, t2;
vector<int> trainID, valID; // indicate which samples are included in the current tree node
char filename[1024];
float *valDist = (float*)malloc(sizeof(float) * classNum); // sample distribution in the val set
for (int i = 0; i < param.treeNum; i++)
{
t1 = clock();
GetFilename(filename);
mexPrintf("\nThe current tree will be in this file: %s\n", filename);
trainID.clear();
valID.clear();
InitializeTree(trees[i]);
int nodeID = 0;
memset(valDist, 0, sizeof(float) * classNum);
TrainDecisionTree(trees[i], set_imVec, param, trainID, valID,
classNum, nodeID, treeMemory, treeMemID, valDist);
// sprintf(filename, "trees/tree_%d%d%d.txt", i / 100, (i % 100) / 10, i % 10);
OutputTree(trees[i], filename);
t2 = clock();
mexPrintf("\n Time for trainning this tree: %f\n", (float)(t2 - t1) / CLOCKS_PER_SEC);
}
free(valDist);
for(int i = 0 ; i < set_imVec.size(); i++)
{
ReleaseData(set_imVec.at(i));
}
for (int i = 0; i < param.treeNum; i++)
ReleaseTreeMem(trees[i]);
delete[] treeMemory;
}
// matlab entry point
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) {
// input must have tree size, binary_feature_filename, binary_feature_flipped_filename, action name
if (nrhs != 4)
mexErrMsgTxt("Wrong number of inputs, treesize, binaryfeaturefilename, binaryfeatureflippedfilename, action_name needed");
// Currently, we do not have any output
if (nlhs != 0)
mexErrMsgTxt("Wrong number of outputs");
/* make sure the first input argument is scalar */
if( !mxIsDouble(prhs[0]) ||
mxIsComplex(prhs[0]) ||
mxGetNumberOfElements(prhs[0])!=1 ) {
mexErrMsgIdAndTxt("MyToolbox:arrayProduct:notScalar",
"Input tree size must be a scalar.");
}
/* input must be a string */
if ( mxIsChar(prhs[1]) != 1 || mxIsChar(prhs[2]) != 1 || mxIsChar(prhs[3]) != 1 )
mexErrMsgIdAndTxt( "MATLAB:revord:inputNotString",
"Input must be a string.");
// read in input string
char *DataFilename, *DataFilenameFlipped, *action;
DataFilename = mxArrayToString(prhs[1]);
DataFilenameFlipped = mxArrayToString(prhs[2]);
action = mxArrayToString(prhs[3]);
// set the seed for random numbers
srand(time(NULL));
mexPrintf("Entered dttrain.cpp!\n"); // debug!!!!
// set parameters for learning
PARAMETER param;
SetParameters(param);
param.treeNum = mxGetScalar(prhs[0]); // treat tree num specially
mexPrintf("tree num is %d\n", param.treeNum); // debug!!!!
mexPrintf("parameters set!\n"); // debug!!!!
// get image feature descriptors from the file
vector<SINGLEIM> imVec;
int classNum;
// use filename passed in here!!!
GetData(imVec, DataFilename, param);
mexPrintf("read %s to image vector! total size of %d\n", DataFilename, imVec.size()); // debug!!!!
//GetData(imVec, DataFilenameFlipped, param);
//mexPrintf("read %s to image vector, total size of %d\n", DataFilenameFlipped, imVec.size()); // debug!!!!
AdjustLabels(imVec, classNum);
mexPrintf("Finish getting data from the disk.\n");
// pre-pooling for the background information, if it is necessary
// if (param.bgPrePooling == 1)
// BgPrePooling(imVec, param);
// mexPrintf("Finish getting background feature.\n");
// the pre-allocated memory for all the trees
int treeMemSize = int(pow(2, param.maxTreeDepth + 1)) * param.treeNum;
TREENODE *treeMemory = new TREENODE[treeMemSize];
InitTreeMem(treeMemory, treeMemSize);
int treeMemID = 0;
vector<TREENODE*> trees; // the set of decision trees
for (int i = 0; i < param.treeNum; i++)
{
trees.push_back(treeMemory + treeMemID);
treeMemID++;
}
mexPrintf("Finish initializing all the trees.\n");
// train decision trees
clock_t t1, t2;
vector<int> trainID, valID; // indicate which samples are included in the current tree node
char filename[1024];
float *valDist = (float*)malloc(sizeof(float) * classNum); // sample distribution in the val set
for (int i = 0; i < param.treeNum; i++)
{
t1 = clock();
GetFilename(filename, action);
mexPrintf("\nThe current tree will be in this file: %s\n", filename);
trainID.clear();
valID.clear();
InitializeTree(trees[i]);
int nodeID = 0;
memset(valDist, 0, sizeof(float) * classNum);
TrainDecisionTree(trees[i], imVec, param, trainID, valID,
classNum, nodeID, treeMemory, treeMemID, valDist);
// sprintf(filename, "trees/tree_%d%d%d.txt", i / 100, (i % 100) / 10, i % 10);
OutputTree(trees[i], filename);
t2 = clock();
mexPrintf("\n Time for trainning this tree: %f\n", (float)(t2 - t1) / CLOCKS_PER_SEC);
}
free(valDist);
ReleaseData(imVec);
for (int i = 0; i < param.treeNum; i++)
ReleaseTreeMem(trees[i]);
delete[] treeMemory;
}
