0
$\begingroup$

In storing floating point values both overflow and underflow problems cause loss of data. In machine learning tasks underflow is a common problem. I wanted to know if double is better than float64 in TensorFlow or not and if there is any difference between them?

$\endgroup$
2
$\begingroup$

Taking a look at tensorflow's dtypes.py, there's this line:

double = float64

So double is exactly the same at float64.

$\endgroup$
  • 1
    $\begingroup$ Might be worth noting that in general (including many C libraries unrelated to question), double tracks what the machine calls double precision (so may depend on processor or OS). Whilst float64 is fixed. There may be a good reason why TensorFlow takes a shortcut here, such as it only works on machines where the two are the same in any case. $\endgroup$ – Neil Slater Nov 30 '17 at 13:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.