I've taken the easy route and given classes of equal score the next integer rank, so class3 and class2 both have rank 2 in sorted_dict1
#!/usr/bin/env python
#Get the ranks for a list of (class, score) tuples sorted by score
#and return them in a dict
def get_ranks(sd):
#The first class in the list has rank 1
k, val = sd[0]
r = 1
rank = {k: r}
for k, v in sd[1:]:
#Only update the rank number if this value is
#greater than the previous
if v > val:
val = v
r += 1
rank[k] = r
return rank
def weighted_mean(a, b):
return (0.50*a + 0.25*b) / 0.75
sorted_dict1 = [('class1', 15.17), ('class2', 15.95), ('class3', 15.95)]
sorted_dict2 = [('class2', 9.10), ('class3', 9.22), ('class1', 10.60)]
print sorted_dict1
print sorted_dict2
ranks1 = get_ranks(sorted_dict1)
ranks2 = get_ranks(sorted_dict2)
print ranks1
print ranks2
keys = sorted(k for k,v in sorted_dict1)
print [(k, weighted_mean(ranks1[k], ranks2[k])) for k in keys]
output
[('class1', 15.17), ('class2', 15.949999999999999), ('class3', 15.949999999999999)]
[('class2', 9.0999999999999996), ('class3', 9.2200000000000006), ('class1', 10.6)]
{'class2': 2, 'class3': 2, 'class1': 1}
{'class2': 1, 'class3': 2, 'class1': 3}
[('class1', 1.6666666666666667), ('class2', 1.6666666666666667), ('class3', 2.0)]
In the comments I mentioned that there's a nice way to create a weighted_mean() function with custom weights. Of course, we could just pass the weights as additional arguments to weighted_mean(), but that makes the call to weighted_mean() more cluttered than it needs to be, making the program harder to read.
The trick is to use a function that takes the custom weights as arguments and returns the desired function. Technically, such a function-making function is called a closure.
Here's a short demo of how to do that.
#!/usr/bin/env python
#Create a weighted mean function with weights w1 & w2
def make_weighted_mean(w1, w2):
wt = float(w1 + w2)
def wm(a, b):
return (w1 * a + w2 * b) / wt
return wm
#Make the weighted mean function
weighted_mean = make_weighted_mean(1, 2)
#Test
print weighted_mean(6, 3)
print weighted_mean(3, 9)
output
4.0
7.0
Here's an updated version of the first program above that handles an arbitrary number of sorted_dict lists. It uses the original get_ranks() function, but it uses a slightly more complex closure than the above example to do the weighted means on a list (or tuple) of data.
#!/usr/bin/env python
''' Weighted means of ranks
From https://stackoverflow.com/q/29413531/4014959
Written by PM 2Ring 2015.04.03
'''
from pprint import pprint
#Create a weighted mean function with weights from list/tuple weights
def make_weighted_mean(weights):
wt = float(sum(weights))
#A function that calculates the weighted mean of values in seq
#weighted by the weights passed to make_weighted_mean()
def wm(seq):
return sum(w * v for w, v in zip(weights, seq)) / wt
return wm
#Get the ranks for a list of (class, score) tuples sorted by score
#and return them in a dict
def get_ranks(sd):
#The first class in the list has rank 1
k, val = sd[0]
r = 1
rank = {k: r}
for k, v in sd[1:]:
#Only update the rank number if this value is
#greater than the previous
if v > val:
val = v
r += 1
rank[k] = r
return rank
#Make the weighted mean function
weights = [0.50, 0.25]
weighted_mean = make_weighted_mean(weights)
#Some test data
sorted_dicts = [
[('class1', 15.17), ('class2', 15.95), ('class3', 15.95), ('class4', 16.0)],
[('class2', 9.10), ('class3', 9.22), ('class1', 10.60), ('class4', 11.0)]
]
print 'Sorted dicts:'
pprint(sorted_dicts, indent=4)
all_ranks = [get_ranks(sd) for sd in sorted_dicts]
print '\nAll ranks:'
pprint(all_ranks, indent=4)
#Get a sorted list of the keys
keys = sorted(k for k,v in sorted_dicts[0])
#print '\nKeys:', keys
means = [(k, weighted_mean([ranks[k] for ranks in all_ranks])) for k in keys]
print '\nWeighted means:'
pprint(means, indent=4)
output
Sorted dicts:
[ [ ('class1', 15.17),
('class2', 15.949999999999999),
('class3', 15.949999999999999),
('class4', 16.0)],
[ ('class2', 9.0999999999999996),
('class3', 9.2200000000000006),
('class1', 10.6),
('class4', 11.0)]]
All ranks:
[ { 'class1': 1, 'class2': 2, 'class3': 2, 'class4': 3},
{ 'class1': 3, 'class2': 1, 'class3': 2, 'class4': 4}]
Weighted means:
[ ('class1', 1.6666666666666667),
('class2', 1.6666666666666667),
('class3', 2.0),
('class4', 3.3333333333333335)]
And here's an alternate version of get_ranks() that skips rank numbers if two or more classes rank the same in a list
def get_ranks(sd):
#The first class in the list has rank 1
k, val = sd[0]
r = 1
rank = {k: r}
#The step size from one rank to the next. Normally
#delta is 1, but it's increased if there are ties.
delta = 1
for k, v in sd[1:]:
#Update the rank number if this value is
#greater than the previous.
if v > val:
val = v
r += delta
delta = 1
#Otherwise, update delta
else:
delta += 1
rank[k] = r
return rank
Here's the output of the program using that alternate version of get_ranks():
Sorted dicts:
[ [ ('class1', 15.17),
('class2', 15.949999999999999),
('class3', 15.949999999999999),
('class4', 16.0)],
[ ('class2', 9.0999999999999996),
('class3', 9.2200000000000006),
('class1', 10.6),
('class4', 11.0)]]
All ranks:
[ { 'class1': 1, 'class2': 2, 'class3': 2, 'class4': 4},
{ 'class1': 3, 'class2': 1, 'class3': 2, 'class4': 4}]
Weighted means:
[ ('class1', 1.6666666666666667),
('class2', 1.6666666666666667),
('class3', 2.0),
('class4', 4.0)]
