Ilya Razenshteyn

there are lots of NaN's in the distance matrix for the example dataset

1

When I run the example script inside VMWare with Ubuntu installed as a guest OS, I get a matrix with around 100K NaN entries. Could it be a problem with...

Are numerical results available as well?

5

Right now, I can see only plots in the repo, which are pretty useless for the "external" comparison as such. In particular, is HNSW still using hours for preprocessing? :)...

See https://github.com/FALCONN-LIB/FFHT/issues/17

Slow and memory-costly compilation

2

Currently, compiling anything that includes FALCONN takes quite a bit of time and RAM (on towhee running `make python_package_install` eats more than 2.4 Gb of RAM). It's a bit painful:...

Travis CI environment seems to have the macros __AVX__ defined, but the registers `ymm??` are not there.

pdoc does not really do a good job.

Currently, it is done by hand via `clang-format --style=Google -i ` for C++ and `yapf -i ` for Python.

Python wrapper for the sparse case

1

Currently, the Python wrapper supports only dense data (via Numpy arrays), would be nice to generalize it to handle sparse data as well.

Optimize probing

1

In particular, try radix sort instead of `std::sort`.