0

I tried to word the question as simply as possible but I'm new to Python and very bad at logic so I'm having a bit of trouble. Basically I want to know if there's a cleaner way to count confusion matrices of two 1D arrays of booleans.

Here's an example:

arr1 = [0, 0, 1, 0, 1, 1]
arr2 = [1, 0, 0, 0, 1, 0]
tp = fp = tn = fn = 0

for i,p in enumerate(arr1):

    a = arr2[i]

    if  p &  a: tp += 1
    if  p & ~a: fp += 1
    if ~p & ~a: tn += 1
    if ~p &  a: fn += 1
    # This was pointed out to be incorrect (see Mozway's answer below)

I tried this but it just adds more lines and looks arguably worse:

if p == o:
    if p: tp += 1
    else: tn += 1
else:
    if p: fp += 1
    else: fn += 1

I then tried adding nested conditional expressions (I believe these are Python's version of ternary operators?) but this disgusting monstrosity doesn't even compile:

(tp += 1 if a else fp += 1) if p else (tn += 1 if ~a else fn += 1)

Any help would be appreciated!

EDIT: Sorry I should have clarified, the result I want is this:

Adding print(tp, fp, tn, fn) would give 1, 2, 2, 1. Simply counting the combinations of each of the booleans in the arrays.

Kentalian
  • 25
  • 3

1 Answers1

2

Use zip and collections.Counter:

from collections import Counter

c = Counter(zip(arr1, arr2))

tp = c[1,1]
fp = c[1,0]
tn = c[0,0]
fn = c[0,1]

print(tp, fp, tn, fn)

output: (1, 2, 2, 1)

counter:

print(c)
# Counter({(0, 1): 1, (0, 0): 2, (1, 0): 2, (1, 1): 1})
alternative way to index the counter:
ids = {'tp': (1,1), 'fp': (1,0), 'tn': (0,0), 'fn': (0,1)}

c[ids['tp']]
# 1
why your first approach failed

bool(~1) is True (~1 is -2), thus giving incorrect counts. ~ is used as not in a vectorial setup (e.g. with numpy), but not in pure python. You can use 1-x (not ~x) to invert an "integer as boolean" (1-0 -> 1 ; 1-1 -> 0).

References on the ~ and & operators in pure python (you should use not and and). Their meaning is different in numpy.

mozway
  • 194,879
  • 13
  • 39
  • 75