python sorted function with user defined cmp functions

Question

I want to order the items in a dictionary using various comparator functions. Please see my example code below. It is the last part using cmpRatio function with sorted() that does not work. I am not sure what I am doing wrong. Thanks in advance for any idea!

mydict = { 'a1': (1,6),
          'a2': (10,2),
          'a3': (5,3),
          'a4': (1,2),
          'a5': (3,9),
          'a6': (9,7) }

# sort by first element of the value tuple: WORKS
print sorted(mydict.iteritems(), key=lambda (k,v): v[0])

# sort by second element of the value tuple: WORKS
print sorted(mydict.iteritems(), key=lambda (k,v): v[1])

# THIS is what I can't get working:
def cmpRatio(x,y):
   sx = float(x[0])/x[1]
   sy = float(y[0])/y[1]
   return sx < sy

# sort by sum of the elements in the value tuple: DOES NOT WORK
print sorted(mydict.iteritems(), key=lambda (k,v): v, cmp=cmpRatio)

Answer 1

Avoid cmp functions where possible because they are slow. They have to be re-evaluated for each comparison. Using a key makes it so the key only needs to be computed once.

print sorted(mydict.iteritems(), key=lambda (k,v): float(v[0])/v[1])

Also, you say you want to sort by the sum of the value items, yet you are sorting by the difference. Sum would look like:

print sorted(mydict.iteritems(), key=lambda (k,v): sum(v))

As mentioned in other answers, for the purpose of really wanting to define a cmp function, you are not returning the proper value (must be -1,0, or 1).

return cmp(sx,sy)

But also if you are just using a lambda to get the value, you can replace that with itemgetter which should be faster than a python-side function:

from operator import itemgetter

print sorted(mydict.iteritems(), key=itemgetter(1), cmp=cmpRatio)

If you are trying to store up sort operations, it would be much better to store the key functions:

key_ops = {
    'sum': lambda (k,v): sum(v),
    'ratio': lambda (k,v): float(v[0])/v[1]),
}

def print_op(aDict, opName):
    print sorted(aDict.iteritems(), key=key_ops[opName])

... # some place later in code
print_op(mydict, 'sum')

Answer 2

you comparison function should return a (negative / zero / positive) value when the first argument is (less than / equal to / greater than) the second value (unlike a comparator given to std::sort(...) in C++).

i.e. instead of

return sx < sy

do

return cmp(sx,sy)