def test_array_match_wrong():
stat = gold.RightWrongCount()
stat.compare_lists = True
gold_ans = {'hello': [1, 9, 3]}
taskrun = {'hello': [1, 2, 3]}
stat = gold.compute(stat, taskrun, gold_ans, 'hello')
stat.right == 0
stat.wrong == 1
def test_confusion_matrix_list_2():
stat = gold.ConfusionMatrix(['True', 'False'])
gold_ans = {'hello': ['True', 'False', 'False', 'False']}
taskrun = {'hello': ['True', 'True', 'False', 'True']}
stat = gold.compute(stat, taskrun, gold_ans, 'hello')
assert stat.value['matrix'][0][0] == 1, stat.value['matrix']
assert stat.value['matrix'][0][1] == 0, stat.value['matrix']
assert stat.value['matrix'][1][0] == 2, stat.value['matrix']
assert stat.value['matrix'][1][1] == 1, stat.value['matrix']
def test_many_labels():
stat = gold.ConfusionMatrix(['A', 'B', 'C', 'D'])
true = ['A', 'A', 'B', 'B', 'C', 'C', 'D', 'D']
obs = ['B', 'C', 'B', 'B', 'D', 'C', 'A', 'B']
gold_ans = {'hello': [{'world': value} for value in true]}
taskrun = {'hello': [{'world': value} for value in obs]}
stat = gold.compute(stat, taskrun, gold_ans, 'hello.world')
assert stat.value['matrix'][0][1] == 1, stat.value['matrix']
assert stat.value['matrix'][0][2] == 1, stat.value['matrix']
assert stat.value['matrix'][1][1] == 2, stat.value['matrix']
assert stat.value['matrix'][2][2] == 1, stat.value['matrix']
assert stat.value['matrix'][2][3] == 1, stat.value['matrix']
assert stat.value['matrix'][3][0] == 1, stat.value['matrix']
assert stat.value['matrix'][3][1] == 1, stat.value['matrix']
assert (np.array(stat.value['matrix']) >= 0).all(), stat.value['matrix']
assert np.array(stat.value['matrix']).sum() == len(true)
def test_confusion_matrix_invalueid_answer():
stat = gold.ConfusionMatrix(['True', 'False'])
gold_ans = {'hello': 'True'}
taskrun = {'hello': 'None'}
stat = gold.compute(stat, taskrun, gold_ans, 'hello')
stat.value['matrix'] == [[0, 0], [0, 0]]
