float precision comparison by specific digit

Question

How to test if two floats are identical until a specific digit?

I tried,

aa1 = 0.043403
aa2 = 0.043392
print(int(aa1*1000) == int(aa2*1000))
>> True

I want to follow this way, but my data include NAN value, it cannot convert it to intro anyhow. I also tried math.isclose but it's tricky.

For example, I wanted to keep until 3 digits and applied the math.isclose

aa3 = 0.013041
aa4 = 0.012545
aa6 = 0.012945

print(math.isclose(aa3, aa4, abs_tol = 0.0001))
>>Flase
print(math.isclose(aa3, aa5, abs_tol = 0.0001))
>>True

But I want to get False for both cases.

Any simple idea??

Do you want to test the rounded values of the numbers up to 3 digits or just a cut after 3 digits? — Jonas, Jun 30 '20 at 12:01
Do you really care about digit? Else "rel_tol" is close enough, but for decimal digits, you need to convert to decimal, so you should convert to string and compare strings [take care about lenght, dot, exponent,..] — Giacomo Catenazzi, Jun 30 '20 at 12:30
Judging from your comments to the other answers I think you have a fundamental misunderstanding of python floating point types. While not mandated, you'll find that every implementation of python uses a IEEE 754 binary floating point type. Those types do not have base 10 digits, they have base 2 digits. Thus there is no such value as .1 in one of these floating point types. Your comment below "... 0.0014, 0.001399, threshold = 0.00001. Should be False..." is problematic because neither one of those values exists in a floating point type. Suggest you consider the `decimal` module. — President James K. Polk, Jun 30 '20 at 13:18
@PresidentJamesK.Polk I totally didn't know there is a concept behind the Python float. I will figure it out later — Dong-gyun Kim, Jun 30 '20 at 13:23

nagyl · Accepted Answer · 2020-06-30T12:52:06.133

3

Create a function which takes 2 numbers, a and b, and it has a preset threshold value. If the difference between a and b is less than the treshold, return True.

def isSame(a, b):
    return abs(a - b) < treshold

Where treshold can be a float. 0.0001 will ignore after the 3rd digit. Or cut the number as a string after 3 digits, and convert it back to float.

Edit, cut off method:

def isSame(a, b, digit):
    return float(str(a)[0:digit+2]) == float(str(b)[0:digit+2])

We ignore every number after the 3rd digit, in case your digit = 3.

edited Jun 30 '20 at 12:52

answered Jun 30 '20 at 12:05

nagyl

1,644
1
7
18

I may miscalculated the treshold value. But you get the point. – nagyl Jun 30 '20 at 12:23
well that was my question :/ – Dong-gyun Kim Jun 30 '20 at 12:24
Works for me. You mean 3 digits after the floating point, or with the leading 0 before the floating point? – nagyl Jun 30 '20 at 12:29
Oh sorry, try this 0.0014, 0.001399, threshold = 0.00001. Should be False, but return True – Dong-gyun Kim Jun 30 '20 at 12:30
I see your problem. This function says True, because basically those numbers are the same. What you need is cutting off the numbers after the 3rd digit. I'll post an update in a minute. – nagyl Jun 30 '20 at 12:48

score 2 · Answer 2 · answered Jun 30 '20 at 12:05

2

You can do it easily using assertAlmostEqual Below is the code

import unittest

unittest.TestCase.assertAlmostEqual(value1,valu2,decimalUpto)

In your case it can be like

unittest.TestCase.assertAlmostEqual(aa1 ,aa2 ,3)

Good Luck

answered Jun 30 '20 at 12:05

Imran

775
7
19

It brings an error AttributeError: 'float' object has no attribute '_formatMessage' – Dong-gyun Kim Jun 30 '20 at 12:08
You have to use it in class .. There are many examples [here](https://programtalk.com/python-examples/unittest.TestCase.assertAlmostEqual/) Have a look on them – Imran Jun 30 '20 at 12:18

score 0 · Answer 3 · answered Jun 30 '20 at 13:13

rel_tol should be enough:

import math

def check(a, b):
    print(math.isclose(a, b, rel_tol=0.001), abs(a/b - 1.) < 0.001)

check(aa1, aa2)
check(aa3, aa4)
check(aa3, aa6)

gives the expected

True True
False False
False False

Note: the second calculation do not handle two equal zeros: easy to add. Not sure if underflow will affect it.

float precision comparison by specific digit

3 Answers3