def test__GroupProbaScorer():
np.random.seed(123)
X = np.random.randn(100, 10)
y = 1 * (np.random.randn(100) > 0)
groups = np.array([0] * 25 + [1] * 25 + [2] * 25 + [3] * 25)
logit = LogisticRegression(solver="lbfgs", random_state=123)
logit.fit(X, y)
scorer = _GroupProbaScorer(score_func=max_proba_group_accuracy,
sign=1,
kwargs={})
res = scorer(logit, X, y, groups)
assert isinstance(res, float)
assert 0 <= res <= 1
assert not pd.isnull(res)
with pytest.raises(TypeError):
res = scorer(logit, X, y) # should not work because group is missing
def test_log_loss_patched_multioutput():
np.random.seed(123)
X = np.random.randn(100, 2)
y1 = np.array(["AA"] * 33 + ["BB"] * 33 + ["CC"] * 33 + ["DD"])
y2 = np.array(["aaa"] * 50+ ["bbb"] * 40 + ["ccc"]* 9 + ["ddd"])
y2d = np.concatenate((y1[:,np.newaxis],y2[:,np.newaxis]),axis=1)
clf = RandomForestClassifier(n_estimators=10,random_state=123)
clf.fit(X,y2d)
scorer = log_loss_scorer_patched()
s = scorer(clf,X,y2d)
assert isinstance(s, float) # verify that the scorer works
y_pred = clf.predict_proba(X)
s2 = -0.5*log_loss(y2d[:,0],y_pred[0]) -0.5*log_loss(y2d[:,1],y_pred[1])
assert s == s2