I was looking at the problem Fastest way to rank items with multiple values and weightings and came up with the following solution, but with two remaining issues:

import numpy as np# set up values
keys = np.array([['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()# crunch the numbers
res = values * weights# combine results with labels
items = np.hstack((np.array(res), keys))# !First problem - .hstack has promoted the first column from float64 to S4:
# array([['130.', 'key1'],
#        ['230.', 'key2'],
#        ['330.', 'key3']], 
#       dtype='|S4')
# How can I force it to stay numeric?items.sort(reverse=True)   # doesn't work, no 'reverse' argument# !Second problem - how to sort the array in descending order?

You could merge res and keys into a structured array:

import numpy.lib.recfunctions as recfunctions
items = recfunctions.merge_arrays([res,keys])

Since np.sort does not have a reverse=True flag, I think the best you can do is reverse the returned array, (e.g. items[::-1]) or else take the negative of res:

import numpy as np
import numpy.lib.recfunctions as recfunctions# set up values
keys = np.array([['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()# crunch the numbers
res = values * weights# combine results with labels
res = np.asarray(-res)
items = recfunctions.merge_arrays([res,keys])
items.dtype.names = ['res', 'key']
items.sort(order=['res'])
print(items)

yields

[(-330.0, 'key3') (-230.0, 'key2') (-130.0, 'key1')]

Note that refunctions.merge_arrays is just a Python convenience function. It uses zip and np.fromiter. It would definitely be faster to avoid joining res and keys and instead use argsort to find the indices that sort res and use those to reorder keys:

import numpy as np# set up values
keys = np.array([['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()# crunch the numbers
res = values * weights# combine results with labels
res = np.squeeze(np.asarray(res))
idx = np.argsort(res)[::-1]
print(keys[idx])
print(res[idx])

yields

[['key3']['key2']['key1']]
[ 330.  230.  130.]