Illustration of P/R/F1 averaging methods

averaging.py

from __future__ import print_function
import numpy as np
from sklearn.metrics import precision_recall_fscore_support as prfs, confusion_matrix
from sklearn.preprocessing import label_binarize
true = [0, 0, 0, 1, 1, 2]
preds = [('under-generate 1', [0, 0, 0, 0, 1, 2]),
         ('under-generate 2', [0, 0, 0, 1, 1, 0]),
         ('over-generate 1', [0, 1, 1, 1, 1, 2]),
         ('confuse 1 and 2', [0, 0, 0, 1, 2, 1])]

fmt = lambda x: '%0.2f' % x

classes = np.unique(true)
for name, pred in preds:
    print(name)
    print(confusion_matrix(true, pred))
    p, r, f, _ = prfs(true, pred, average=None)
    for i in classes:
        print(fmt(p[i]), fmt(r[i]), fmt(f[i]), 'class %d' % i, sep='\t')
    for average in ['macro', 'micro', 'weighted']:
        p, r, f, _ = prfs(true, pred, average=average)
        print(fmt(p), fmt(r), fmt(f), average, sep='\t')
        
    true_ml = label_binarize(true, classes)[:, 1:]
    pred_ml = label_binarize(pred, classes)[:, 1:]
    for average in ['macro', 'micro', 'weighted']:
        p, r, f, _ = prfs(true_ml, pred_ml, average=average)
        print(fmt(p), fmt(r), fmt(f), average + ' minor', sep='\t')
    print()

averaging.txt

under-generate 1
[[3 0 0]
 [1 1 0]
 [0 0 1]]
0.75	1.00	0.86	class 0
1.00	0.50	0.67	class 1
1.00	1.00	1.00	class 2
0.92	0.83	0.84	macro
0.83	0.83	0.83	micro
0.88	0.83	0.82	weighted
1.00	0.75	0.83	macro minor
1.00	0.67	0.80	micro minor
1.00	0.67	0.78	weighted minor

under-generate 2
[[3 0 0]
 [0 2 0]
 [1 0 0]]
0.75	1.00	0.86	class 0
1.00	1.00	1.00	class 1
0.00	0.00	0.00	class 2
0.58	0.67	0.62	macro
0.83	0.83	0.83	micro
0.71	0.83	0.76	weighted
0.50	0.50	0.50	macro minor
1.00	0.67	0.80	micro minor
0.67	0.67	0.67	weighted minor

over-generate 1
[[1 2 0]
 [0 2 0]
 [0 0 1]]
1.00	0.33	0.50	class 0
0.50	1.00	0.67	class 1
1.00	1.00	1.00	class 2
0.83	0.78	0.72	macro
0.67	0.67	0.67	micro
0.83	0.67	0.64	weighted
0.75	1.00	0.83	macro minor
0.60	1.00	0.75	micro minor
0.67	1.00	0.78	weighted minor

confuse 1 and 2
[[3 0 0]
 [0 1 1]
 [0 1 0]]
1.00	1.00	1.00	class 0
0.50	0.50	0.50	class 1
0.00	0.00	0.00	class 2
0.50	0.50	0.50	macro
0.67	0.67	0.67	micro
0.67	0.67	0.67	weighted
0.25	0.25	0.25	macro minor
0.33	0.33	0.33	micro minor
0.33	0.33	0.33	weighted minor