Open machow opened 2 months ago
It seems that numpy
is also utilized in some tests. Should we consider removing it from testing as well?
It seems okay to keep it around in tests. Since pandas can wrap numpy, and return numpy typed objects, we might need it in tests to validate that functions do the right things for a few numpy objects (which might make it in via pandas, etc..)
Currently, we use numpy in two places:
_tbl_data.py
: this is where we've consolidated all our DataFrame library specific code. It's handled using databackend, so we don't need to depend on these libraries to do isinstance checks on them, etc..._utils_nanoplots.py
,_data_color/base.py
. This uses numpy in two ways._tbl_data.py
np.linspace()
, andnp.mean()
should be re-implemented internally. We don't use too many numpy functions, so it hopefully isn't too nuts to do.Note that for replacing numpy in performing calculations, we should be careful to ensure tests are in place / added where needed. We have good coverage, but I'm not sure it covers every numpy path.