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.
I guess you can always just add an assert not data.isna().any() in strategic locations
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.
this is just like in regular math too. not being a number is just so fun that nobody wants to go back to being a number once they get a taste of it
Fucking over-dramatic divisions by 0, sigh.
The funniest thing about NaNs is that they’re actually coded so you can see what caused it if you look at the binary. Only problem is; due to the nature of NaNs, that code is almost always going to resolve to “tried to perform arithmetic on a NaN”
There are also coded NaNs which are defined and sometimes useful, such as +/-INF, MAX, MIN (epsilon), and Imaginary
Thanks. This is great
