Kajika comments on Floating-point arithmetic

[–] Kajika@lemmy.ml [S] 80 points 2 years ago (2 children)

Took me 2 hours to find out why the final output of a neural network was a bunch of NaN. This is always very annoying but I can't really complain, it make sense. Just sucks.

permalink

fedilink

source

hideshow 4 child comments

[–] kurwa@lemmy.world 44 points 2 years ago

I hope it was garlic NaN at least.

permalink

fedilink

source

parent

[–] flying_sheep@lemmy.ml 13 points 2 years ago (2 children)

I guess you can always just add an assert not data.isna().any() in strategic locations

permalink

fedilink

source

parent

hideshow 4 child comments

[–] Kajika@lemmy.ml [S] 29 points 2 years ago (4 children)

That could be a nice way. Sadly it was in a C++ code base (using tensorflow). Therefore no such nice things (would be slow too). I skill-issued myself thinking a struct would be 0 -initialized but MyStruct input; would not while MyStruct input {}; will (that was the fix). Long story.

permalink

fedilink

source

parent

hideshow 8 child comments

[–] fkn@lemmy.world 12 points 2 years ago

I too have forgotten to memset my structs in c++ tensorflow after prototyping in python.

permalink

fedilink

source

parent

[–] TheFadingOne@feddit.de 5 points 2 years ago*

If you use the GNU libc the feenableexcept function, which you can use to enable certain floating point exceptions, could be useful to catch unexpected/unwanted NaNs

permalink

fedilink

source

parent

[–] nickwitha_k@lemmy.sdf.org 1 point 2 years ago

Oof. This makes me appreciate the abstractions in Go. It's a small thing but initializing structs with zero values by default is nice.

permalink

fedilink

source

parent

[–] flying_sheep@lemmy.ml 1 point 2 years ago

Oof. C++ really is a harsh mistress.