search

eddelbuettel / rcppcnpy

Rcpp bindings for NumPy files
GNU General Public License v2.0
26 stars 16 forks source link

npyLoad crashes or shows wrong values if array is not Python type float64 #29

Closed JaySLee closed 4 years ago

JaySLee commented 4 years ago

Hi again,

I might as well mention this as well. If the numpy array is not set to float64 (e.g., float32 or int64), npyLoad() either crashes or shows incorrect values. I haven't looked further into this, but if you want me to suggest changes to the documentation for now, I can do a pull request for that.

Jay

eddelbuettel commented 4 years ago

Well the test code in tests/loadFiles.R is clearly a counter example as it uses int (with the required and documented argument).

JaySLee commented 4 years ago

Whoops, I didn't read the documentation closely. My bad!

eddelbuettel commented 4 years ago

Ahh, that happens. I'll close this. Feel free to re-open if you find something amiss.

mmfalco commented 3 years ago

Hi @eddelbuettel, is there a way to read directly numpy boolean arrays with your package? Not sure if I should open a new issue with the request, but I wanted to ask first in case I'm missing something.

eddelbuettel commented 3 years ago

I don't know, I have not tried.

It could, the function cnpy::map_type() (in file src/cnpy.cpp, and from the cnpy library implementing the actual functionality of reading NumPy data) appears to have a mapper for boolean type in the NumPy header. That should work for a file written by Python. Our initial focus years ago was just float(64) and int(32) plus "neighbouring" types. One issue may be that R has three values for a boolean (!!), namely TRUE, FALSE and NA, whereas Python only has two. But you might get TRUE and FALSE in.

These days you can also rely on reticulate (if you have it installed) for a much heavier and powerful converter for all things Python (including data); I show in the second vignette how to use that.