Ralf Meermeier
Ralf Meermeier
Is there any resolution to this issue planned? The current behavior maps 40% of the dynamic range of a float (10^-6 to 10^-38) to one single value, 0.0. For doubles...
Just a quick note from my side, for the purposes of Tensorflow model conversion, a simple switch to std::scientific in FloatToString() did the trick and fixed all my problems while...
Was this PR (and Simon's) reviewed to death? It's a shame it never went in, this is an incredibly useful tool.