getting mantissa size of float

[Skaperen]

i know the computational way to determine the number of bits of mantissa of float (have done it in C). i am wondering if there is some defined constant(s) or other way to get this more directly (across various platforms). i need to limit this to modules that come built in to Python. i want the effective mantissa (53 for x86 double) rather than the reserved space (52 for x86 double).

[DeaD_EyE]

Not so easy. From stackoverflow: https://stackoverflow.com/questions/4533...ut-strings

I think this is the best solution (float.mantissa() and float.exponent() would be nicer):

from decimal import Decimal


def fexp(number):
    (sign, digits, exponent) = Decimal(number).as_tuple()
    return len(digits) + exponent - 1


def fman(number):
    return Decimal(number).scaleb(-fexp(number)).normalize()

Or for easier use a class:

from decimal import Decimal


class MyFloat(float):

    @property
    def fexp(self):
        (sign, digits, exponent) = Decimal(self).as_tuple()
        return len(digits) + exponent - 1

    @property
    def fman(self):
        return Decimal(self).scaleb(-self.fexp).normalize()


v = MyFloat(133.7)
print(v.fexp)
print(v.fman)

Output:
2
1.336999999999999886313162278

[Gribouillis]

i want the effective mantissa (53 for x86 double) rather than the reserved space (52 for x86 double).

What does this mean? What is the «effective mantissa». Do you have a reference for this?

[DeaD_EyE]

There is a function in the module math:
https://docs.python.org/3/library/math.html#math.frexp

It returns different values, but after the calculation you get the same result.
I'm not complete sure which one is the right representation.

[DeaD_EyE]

What does this mean? What is the «effective mantissa». Do you have a reference for this?

IEEE 754: A double precision float (64 bit) has 52 bits for mantissa. The 53th bit is implicit and always True. It's a trick from old men with long beards.

[Gribouillis]

I dont agree with that. There is 1 sign bit, 11 bits of exponent and 52 bits of mantissa. That makes 64 bits. Or perhaps you are talking about the implicit 1. in normal numbers? I wrote a gist about IEEE754. This doesn't make Skaperen's question more clear. I think he means something else.

Perhaps sys.float_info.mant_dig is the required information? It gives 53 on my system.

[DeaD_EyE]

Yes, I was talking about the implicit 1.

Perhaps sys.float_info.mant_dig is the required information? It gives 53 on my system.

Same here. It's hard to find a different result.

[Skaperen]

Perhaps sys.float_info.mant_dig is the required information? It gives 53 on my system.

that's what i'm looking for. i was searching "mantissa" but that word was not used so it was not found. a slow read about the float type would have gotten the reference but, as usual, i am in a hurry and trying to find things the most direct way.