OpenCV 4.5.5 how to use findNearest - c++

Question

I am using OpenCv 4.5.5 and trying to perform NN search using the 'findNearest' provided by ml::KNearest, the classifier has been previously trained and saved to disk and I load it before attempting to perform the search.
I checked that it is properly saved and loaded, and that the types are always CV_32F so I dont think the issue is there. The dimensions of the testing samples is the same as the ones used in training. During runtime I get the following error:
kdtree.cpp:279: error: (-215:Assertion failed) vecmat.isContinuous()
&& vecmat.type() == CV_32F && vecmat.total() == (size_t)points.cols in
function 'findNearest'
I get the same error if I try the method 'predict'.
Why in the documentation says that the 'samples' (input) should be a matrix of dimensions <number_of_samples> * k? I would expect to have 'k' times the number of samples
for the results but not for the input.
Here is the relevant code:
void EigenClassifier::train(const cv::Mat& samples, const cv::Mat& labels)
{
auto prototypes = prepare_samples(samples);
std::cout << "EigenClassifier: Building KD-Tree..." <<std::endl;
m_knearest->train(prototypes, cv::ml::ROW_SAMPLE, labels);
std::cout << "Is trained: " << m_knearest->isTrained() << std::endl;
}
cv::Mat EigenClassifier::find_nearest(const cv::Mat& test_samples, int k, cv::Mat& neighbor_responses, cv::Mat& distances)
{
cv::Mat result;
auto reduced_samples = prepare_samples(test_samples);
std::cout << "Reduced samples size: " << reduced_samples.size << " type: " << reduced_samples.type() << std::endl;
m_knearest->findNearest(reduced_samples, k, result, neighbor_responses, distances);
//m_knearest->predict(reduced_samples.row(0));
return result;
}
Do I need to initialize 'result', 'neighbor_responses' and 'distances' to the expected results size?