FluidUMAP: Crash if number input points < numNeighbours + 1

[tremblap]

this patch with this soundfile crash max when at point 2.5 (umap) -there must be something in the points that triggers an exception. Both in Test home compile and in Release CI compile.
Archive.zip

[tremblap]

bug report in Test mode:

Process: Max [1475]
Path: /Applications/Max.app/Contents/MacOS/Max
Identifier: com.cycling74.Max
Version: 8.2.1 [2f7eaf31afd] (8.2.1)
Code Type: X86-64 (Native)
Parent Process: ??? [1]
Responsible: Max [1475]
User ID: 501

Date/Time: 2022-07-10 14:20:03.533 +0200
OS Version: macOS 11.2.1 (20D74)
Report Version: 12
Anonymous UUID: 1CEE2877-ACA1-AF56-6B62-4814F29E546A

Time Awake Since Boot: 720 seconds

System Integrity Protection: enabled

Crashed Thread: 0 CrBrowserMain Dispatch queue: com.apple.main-thread

Exception Type: EXC_CRASH (SIGABRT)
Exception Codes: 0x0000000000000000, 0x0000000000000000
Exception Note: EXC_CORPSE_NOTIFY

Application Specific Information:
Assertion failed: (impl::checkBounds(mDesc, args...) && "Arguments out of bounds"), function operator(), file /Users/pa/Documents/documents@hudd/research/projects/fluid corpus navigation/research/fluid_decomposition/include/clients/nrt/../common/../../data/FluidTensor.hpp, line 568.

Thread 0 Crashed:: CrBrowserMain Dispatch queue: com.apple.main-thread
0 libsystem_kernel.dylib 0x00007fff2047a462 __pthread_kill + 10
1 libsystem_pthread.dylib 0x00007fff204a8610 pthread_kill + 263
2 libsystem_c.dylib 0x00007fff203fb720 abort + 120
3 libsystem_c.dylib 0x00007fff203fa9d6 __assert_rtn + 314
4 org.flucoma.fluid.libmanipulation 0x0000000134592c92 fluid::algorithm::UMAP::makeGraph(fluid::FluidDataSet<std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator >, double, 1l> const&, long, Eigen::SparseMatrix<double, 0, int>&, Eigen::Ref<Eigen::Array<double, -1, -1, 0, -1, -1>, 0, Eigen::OuterStride<-1> >, bool) const + 946
5 org.flucoma.fluid.libmanipulation 0x00000001345907bd fluid::algorithm::UMAP::train(fluid::FluidDataSet<std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator >, double, 1l>&, long, long, double, long, double) + 1277
6 org.flucoma.fluid.libmanipulation 0x000000013458c58f fluid::client::umap::UMAPClient::fitTransform(fluid::client::SharedClientRef<fluid::client::dataset::DataSetClient const>, fluid::client::SharedClientReffluid::client::dataset::DataSetClient) + 1039

[weefuzzy]

Think this is a duplicate of #104 Actually, no. Same basic problem (not enough points) but different code path.

[weefuzzy]

In makeGraph we can see that the code assumes it can rely on there being at least numNeighbours + 1 data points if discardFirst is true.

@g-roma I'm assuming the thing to do here, as with #104, is to add a check into UMAPClient that we actually have enough data points to proceed, and maybe also an assertion here in makeGraph ?

flucoma-core/include/algorithms/public/UMAP.hpp

Lines 265 to 283 in 804a3b3

	void makeGraph(const DataSet& in, index k, SparseMatrixXd& graph,
	Ref<ArrayXXd> dists, bool discardFirst) const
	{
	graph.reserve(in.size() * k);
	auto data = in.getData();
	for (index i = 0; i < in.size(); i++)
	{
	auto nearest = mTree.kNearest(data.row(i), discardFirst ? k + 1 : k);
	auto nearestIds = nearest.getIds();
	auto distances = nearest.getData().col(0);
	for (index j = 0; j < k; j++)
	{
	index pos = discardFirst ? j + 1 : j;
	index neighborIndex = stoi(nearestIds(pos));
	dists(i, j) = distances(pos);
	graph.insert(i, neighborIndex) = distances(pos);
	}
	}
	}