检索引起运行时警告的数组元素

Question

When I apply numpy.exp to an array of numbers I get the following runtime warning:

RuntimeWarning: overflow encountered in exp

Now I know that some value in this array caused the result of exp to overflow but I don't know which one (and this array has millions of entries).

Is there a way to make this runtime warning more verbose in a sense that it also displays the specific argument that caused the overflow (and more importantly its position in the input array)? I discovered seterr and seterrcall however they don't seem to incorporate what caused the error (only what the error is about).

---

I know that - after encountering this warning - I could pass each value in the array separately to exp and watch out for the warning or that I could use isfinite to test the elements of exp(array) . However this implies messing with the actual code while I would prefer that numpy does this kind of things behind the scenes (being configured appropriately).

---

Edit: I was asked to post some code that reproduces the error. Here it is:

>>> import numpy
>>> array = numpy.zeros((10,), dtype=float)
>>> array[0] = 1.0e308
>>> numpy.exp(array)
__main__:1: RuntimeWarning: overflow encountered in exp
array([ inf,   1.,   1.,   1.,   1.,   1.,   1.,   1.,   1.,   1.])

I would like to know that it was the element with index 0 that caused to overflow (not just that an overflow happened somewhere).

Answer 1

The problem is just that exp(1e308) is too big for float !

If you consider the exp function, knowing that 1.7976931348623157e+308 is the maximum value for a float, remove from your array all the value below log(1.7976931348623157e+308) = 709