I have a collection of key value pairs like this:
{
\'key1\': [value1_1, value2_1, value3_1, ...],
\'key2\': [value1_2, value2_2, value3_2, ...],
...
You can't get linear-time, but you can do it faster; this looks like a matrix-multiply to me, so I suggest you use numpy
:
import numpy as np
keys = ['key1', 'key2', 'key3']
values = np.matrix([
[1.1, 1.2, 1.3, 1.4],
[2.1, 2.2, 2.3, 2.4],
[3.1, 3.2, 3.3, 3.4]
])
weights = np.matrix([[10., 20., 30., 40.]]).transpose()
res = (values * weights).transpose().tolist()[0]
items = zip(res, keys)
items.sort(reverse=True)
which gives
[(330.0, 'key3'), (230.0, 'key2'), (130.0, 'key1')]
Edit: with thanks to @Ondro for np.dot and to @unutbu for np.argsort, here is an improved version entirely in numpy:
import numpy as np
# set up values
keys = np.array(['key1', 'key2', 'key3'])
values = np.array([
[1.1, 1.2, 1.3, 1.4], # values1_x
[2.1, 2.2, 2.3, 2.4], # values2_x
[3.1, 3.2, 3.3, 3.4] # values3_x
])
weights = np.array([10., 20., 30., 40.])
# crunch the numbers
res = np.dot(values, -weights) # negative of weights!
order = res.argsort(axis=0) # sorting on negative value gives
# same order as reverse-sort; there does
# not seem to be any way to reverse-sort
# directly
sortedkeys = keys[order].tolist()
which results in ['key3', 'key2', 'key1']
.
Here's a normalization function, that will linearly transform your values into [0,1]
def normalize(val, ilow, ihigh, olow, ohigh):
return ((val-ilow) * (ohigh-olow) / (ihigh - ilow)) + olow
Now, use normalize
to compute a new dictionary with normalized values. Then, sort by the weighted sum:
def sort(d, weights, ranges):
# ranges is a list of tuples containing the lower and upper bounds of the corresponding value
newD = {k:[normalize(v,ilow, ihigh, 0, 1) for v,(ilow, ihigh) in zip(vals, ranges)] for k,val in d.iteritems()} # d.items() in python3
return sorted(newD, key=lambda k: sum(v*w for v,w in zip(newD[k], weights)))