Numba with decimal package for increasing the accuracy

Numba Support: How do I do ...?
1

Hi everyone,

Do you know if numba has the format for decimal value type? Or anything else to increase the accuracy of float value?

For example: I have a guvectorize here

guvectorize([(float64[:,:], float64, float64[:,:])], ‘(m,n),()->(m,n)’)

is that possible to change to something like

guvectorize([(float64[:,:], float64, decimal[:,:])], ‘(m,n),()->(m,n)’)

Please help me with solution for this. Thank you.

2

Numba doesn’t presently support decimal types. I would expect that will be possible to add support for it using a Numba extension (see Extending Numba — Numba 0.55.0.dev0+438.g3131959c9.dirty-py3.7-linux-x86_64.egg documentation), but it will require some knowledge / study of how Numba works and the documentation if it’s an area you’ve not looked into before.

1 Like
4

I’m also investigating this as part of my work. Would there be any interest in upstreaming this if that’s the route I go down?