libscran / umappp

C++ port of the UMAP algorithm
https://libscran.github.io/umappp/
BSD 2-Clause "Simplified" License
42 stars 15 forks source link

A C++ library for UMAP

Unit tests Documentation uwot comparison Codecov

Overview

umappp is a header-only C++ implementation of the Uniform Manifold Approximation and Projection (UMAP) algorithm (McInnes, Healy and Melville, 2018). UMAP is a non-linear dimensionality reduction technique that is most commonly used for visualization of complex datasets. This is achieved by placing each observation on a low-dimensional (usually 2D) embedding in a manner that preserves the neighborhood of each observation from the high-dimensional original space. The aim is to ensure that the local structure of the data is faithfully recapitulated in lower dimensions Further theoretical details can be found in the original UMAP documentation; the implementation here is derived from the C++ code in the uwot R package.

Quick start

Given a pointer to a column-major input array with ndim rows and nobs columns, we use initialize() to start the UMAP algorithm and run() to run it across epochs:

#include "umappp/umappp.hpp"

// Set number of dimensions in the output embedding.
size_t out_dim = 2;
std::vector<double> embedding(npts * out_dim);

// Initialize the UMAP state:
umappp::Options opt;
auto status = umappp::initialize(
    ndim,
    nobs,
    data.data(),
    knncolle::VptreeBuilder(), // algorithm to find neighbors
    out_dim,
    embedding.data(),
    opt
);

// Run UMAP algorithm to completion. This modifies the
// values in the output 'embedding' vector.
status.run();

We can modify parameters in the Options class that is passed to initialize():

opt.num_neighbors = 20;
opt.num_epochs = 200;
opt.min_dist = 0.2;

We can also run the algorithm up to the specified number of epochs, which is occasionally useful for inspecting the intermediate states of the embedding:

auto status2 = umappp::initialize(
    ndim,
    nobs,
    data.data(),
    knncolle::VptreeBuilder(), // algorithm to find neighbors
    out_dim,
    embedding.data(),
    opt
);

for (int iter = 10; iter < 200; iter += 10) {
    status2.run(iter);
    // do something with the current embedding, e.g., create an animation
}

Advanced users can control the neighbor search by either providing the search results directly (as a vector of vectors of index-distance pairs) or by providing an appropriate knncolle subclass to the initialize() function:

auto annoy_idx = knncolle::AnnoyBuilder().build_unique(
    knncolle::SimpleMatrix(ndim, nobs, data.data())
);
auto status_annoy = umappp::initialize(*annoy_idx, 2, embedding.data(), opt);
status_annoy.run();

See the reference documentation for more details.

Building projects

CMake with FetchContent

If you're already using CMake, you can add something like this to your CMakeLists.txt:

include(FetchContent)

FetchContent_Declare(
  umappp 
  GIT_REPOSITORY https://github.com/libscran/umappp
  GIT_TAG master # or any version of interest
)

FetchContent_MakeAvailable(umappp)

And then:

# For executables:
target_link_libraries(myexe libscran::umappp)

# For libaries
target_link_libraries(mylib INTERFACE libscran::umappp)

CMake with find_package()

find_package(libscran_umappp CONFIG REQUIRED)
target_link_libraries(mylib INTERFACE libscran::umappp)

To install the library, use:

mkdir build && cd build
cmake .. -DUMAPPP_TESTS=OFF
cmake --build . --target install

By default, this will use FetchContent to fetch all external dependencies. If you want to install them manually, use -DUMAPPP_FETCH_EXTERN=OFF. See extern/CMakeLists.txt to find compatible versions of each dependency.

Manual

If you're not using CMake, the simple approach is to just copy the files in include/ - either directly or with Git submodules - and include their path during compilation with, e.g., GCC's -I. This requires the external dependencies listed in extern/CMakeLists.txt, which also need to be made available during compilation.

References

McInnes L, Healy J, Melville J (2020). UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv, https://arxiv.org/abs/1802.03426