Unexpected Python behavior with add in a class

Question

I wanted to make a probability-like class for practice, so I constructed a class P and wanted to be able to have an associated value with it. I also wanted to be able to add probabilities like P("a") + P("b") and have it add their values. This was fine to code, but I got some strange behavior while testing. I've pasted only the relevant parts of the code below [which is why it might seem a bit verbose]:

class P:

def __init__(self, event):
    self.event = event
    self.v = 0

def value(self, val):
        """Sets the probability to the value 'val'."""
    self.v = val

def add_stuff(x,y):
    return lambda x,y: x+y

def __add__(self, other):

    if isinstance(other, P):   # if we are adding two P's together.
        return add_stuff(self.v, other.v)

    else:                      # if we are adding a number to our P.
        try: return add_stuff(self.v, other)
        except: raise TypeError(self.type_error_string)



a = P("a")  # Creates the instances.
b = P("b")  #
c = P("c")  #

a.value(0.5)  # Sets the value of a.v to 0.5,
b.value(0.1)  # and so on for b and c.
c.value(0.2)  #

print a.v + b.v == 0.7. # prints True.
print b.v == 0.1        # prints True.
print c.v == 0.2        # prints True.
print b.v + c.v         # prints 0.3.
print type(b.v + c.v)   # prints <float>
print b.v + c.v == 0.3  # prints False (!!).

The relevant part here is the bottom. Note that a.v + b.v [as well as some other values] were fine when testing, but not b.v + c.v for some reason. I'm not sure what is happening here.

This has nothing to do with OOP, nothing to do with the `__add__` overload, and everything to do with floating-point arithmetic. — Karl Knechtel, Dec 23 '13 at 00:21
`.2 + .1 == .3` is always false, in pretty much any language using floats — tobyodavies, Dec 23 '13 at 00:22
Dang. Really? I did not know that. I'll delete this question. Is there any way to get around this? — , Dec 23 '13 at 00:22
Thank you guys. I'll flag this for deletion [due to duplication]. If any mods could just delete it, I'd appreciate it! — , Dec 23 '13 at 00:26

BartoszKP · Accepted Answer · 2015-10-23T15:18:51.490

Apart from some other problems with your code, this particular issue lies in the way floating-point arithmetic works:

>>> 0.1 + 0.2
0.30000000000000004
>>>

Consider reading this great article: What Every Computer Scientist Should Know About Floating-Point Arithmetic.

In short: the way to compare floats is introducing some tolerance: abs(b.v - c.v) < Epsilon (you should define Epsilon appropriately, something like 1e-8). You can also use the decimal module.

score 0 · Answer 2 · answered Dec 23 '13 at 00:25

Based on your definitions of add_stuff and __add__, it looks like you need to have this:

def __add__(self,other):
    ...
    return add_stuff(self.v, other.v)(self.v,other.v) # add_stuff() returns a function which gets used in the second set of brackets
    ...

Unexpected Python behavior with __add__ in a class

2 Answers2

Unexpected Python behavior with add in a class