potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 15:23:05 +04:00
Code Packages Releases Wiki Activity

bound categorical split indices in dtrees readSplit

Browse Source
This commit is contained in:
Uwez Khan
2026-07-18 00:57:52 +05:30
committed by Abhishek gola
parent 70fc346580
commit dee2bb288b
2 changed files with 60 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/ml/src/tree.cpp
+5 -1
View File
@@ -1839,12 +1839,14 @@ int DTreesImpl::readSplit( const FileNode& fn )
Split split;
int vi = (int)fn["var"];
CV_Assert( 0 <= vi && vi <= (int)varType.size() );
CV_Assert( 0 <= vi && vi < (int)varMapping.size() );
vi = varMapping[vi]; // convert to varIdx if needed
CV_Assert( 0 <= vi && vi < (int)varType.size() );
split.varIdx = vi;
if( varType[vi] == VAR_CATEGORICAL ) // split on categorical var
{
CV_Assert( vi < (int)catOfs.size() );
int i, val, ssize = getSubsetSize(vi);
split.subsetOfs = (int)subsets.size();
for( i = 0; i < ssize; i++ )
@@ -1860,6 +1862,7 @@ int DTreesImpl::readSplit( const FileNode& fn )
if( fns.isInt() )
{
val = (int)fns;
CV_Assert( 0 <= val && (val >> 5) < ssize );
subset[val >> 5] |= 1 << (val & 31);
}
else
@@ -1869,6 +1872,7 @@ int DTreesImpl::readSplit( const FileNode& fn )
for( i = 0; i < n; i++, ++it )
{
val = (int)*it;
CV_Assert( 0 <= val && (val >> 5) < ssize );
subset[val >> 5] |= 1 << (val & 31);
}
}
modules/ml/test/test_save_load.cpp
+55
View File
@@ -96,6 +96,61 @@ ML_Legacy_Param param_list[] = {
INSTANTIATE_TEST_CASE_P(/**/, ML_Legacy_Params, testing::ValuesIn(param_list));
TEST(ML_DTrees, load_bad_categorical_split)
{
// Train a tree with a categorical input so the model carries a
// categorical split serialized as an "in"/"not_in" value list.
const int n = 40;
Mat samples(n, 1, CV_32F);
Mat responses(n, 1, CV_32S);
for (int i = 0; i < n; i++)
{
int cat = i % 4;
samples.at<float>(i, 0) = (float)cat;
responses.at<int>(i, 0) = (cat == 1 || cat == 2) ? 1 : 0;
}
Mat varType(2, 1, CV_8U);
varType.at<uchar>(0) = ml::VAR_CATEGORICAL;
varType.at<uchar>(1) = ml::VAR_CATEGORICAL;
Ptr<ml::TrainData> td = ml::TrainData::create(samples, ml::ROW_SAMPLE, responses,
noArray(), noArray(), noArray(), varType);
Ptr<ml::DTrees> dt = ml::DTrees::create();
dt->setMaxDepth(4);
dt->setCVFolds(0);
dt->setMaxCategories(4);
dt->setMinSampleCount(1);
dt->train(td);
const string filename = cv::tempfile(".yml");
dt->save(filename);
string model;
{
std::ifstream in(filename.c_str());
std::stringstream ss;
ss << in.rdbuf();
model = ss.str();
}
// Category values in the split list index a per-split subset bitmask.
// Injecting a value larger than the number of categories used to write
// past that bitmask; loading such a model must be rejected, not crash.
size_t pos = model.find("in:");
ASSERT_NE(pos, string::npos);
size_t br = model.find('[', pos);
ASSERT_NE(br, string::npos);
model.insert(br + 1, "1000000,");
{
std::ofstream out(filename.c_str());
out << model;
}
Ptr<ml::DTrees> bad;
EXPECT_THROW(bad = ml::DTrees::load(filename), Exception);
remove(filename.c_str());
}
/*TEST(ML_SVM, throw_exception_when_save_untrained_model)
{
Ptr<cv::ml::SVM> svm;