def test_fit_subsample(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = GradientBoostingSurvivalAnalysis(n_estimators=50, max_features=8, subsample=0.6,
random_state=0)
model.fit(whas500_data.x, whas500_data.y)
assert model.max_features_ == 8
assert hasattr(model, "oob_improvement_")
incl_mask = numpy.ones(whas500_data.x.shape[0], dtype=bool)
incl_mask[[35, 111, 174, 206, 236, 268, 497]] = False
x_test = whas500_data.x[incl_mask]
y_test = whas500_data.y[incl_mask]
p = model.predict(x_test)
assert_cindex_almost_equal(y_test['fstat'], y_test['lenfol'], p,
(0.8330510326740247, 60985, 12221, 2, 14))
assert (50,) == model.train_score_.shape
assert (50,) == model.oob_improvement_.shape
with pytest.raises(ValueError, match="Number of features of the model must match the input. "
"Model n_features is 14 and input n_features is 2 "):
model.predict(whas500_data.x[:, :2])
def test_fit(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = ComponentwiseGradientBoostingSurvivalAnalysis(n_estimators=100)
model.fit(whas500_data.x, whas500_data.y)
p = model.predict(whas500_data.x)
assert_cindex_almost_equal(whas500_data.y['fstat'], whas500_data.y['lenfol'], p,
(0.7755659, 58283, 16866, 0, 14))
expected_coef = pandas.Series(numpy.zeros(15, dtype=float), index=whas500_data.names)
expected_coef['age'] = 0.040919
expected_coef['hr'] = 0.004977
expected_coef['diasbp'] = -0.003407
expected_coef['bmi'] = -0.017938
expected_coef['sho'] = 0.429904
expected_coef['chf'] = 0.508211
assert_array_almost_equal(expected_coef.values, model.coef_)
assert (100,) == model.train_score_.shape
with pytest.raises(ValueError, match='Dimensions of X are inconsistent with training data: '
'expected 14 features, but got 2'):
model.predict(whas500_data.x[:, :2])
def test_fit_subsample(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = ComponentwiseGradientBoostingSurvivalAnalysis(n_estimators=100, subsample=0.6, random_state=0)
model.fit(whas500_data.x, whas500_data.y)
p = model.predict(whas500_data.x)
assert_cindex_almost_equal(whas500_data.y['fstat'], whas500_data.y['lenfol'], p,
(0.7750602, 58245, 16904, 0, 14))
expected_coef = pandas.Series(numpy.zeros(15, dtype=float), index=whas500_data.names)
expected_coef['age'] = 0.041299
expected_coef['hr'] = 0.00487
expected_coef['diasbp'] = -0.003381
expected_coef['bmi'] = -0.017018
expected_coef['sho'] = 0.433685
expected_coef['chf'] = 0.510277
assert_array_almost_equal(expected_coef.values, model.coef_)
assert (100,) == model.train_score_.shape
assert (100,) == model.oob_improvement_.shape
with pytest.raises(ValueError, match='Dimensions of X are inconsistent with training data: '
'expected 14 features, but got 2'):
model.predict(whas500_data.x[:, :2])
def test_compare_clinical_kernel(make_whas500):
whas500 = make_whas500(to_numeric=True)
trans = ClinicalKernelTransform()
trans.fit(whas500.x_data_frame)
kpca = KernelPCA(kernel=trans.pairwise_kernel, copy_X=True)
xt = kpca.fit_transform(whas500.x)
nrsvm = FastSurvivalSVM(optimizer='rbtree',
tol=1e-8,
max_iter=500,
random_state=0)
nrsvm.fit(xt, whas500.y)
rsvm = FastKernelSurvivalSVM(optimizer='rbtree',
kernel=trans.pairwise_kernel,
tol=1e-8,
max_iter=500,
random_state=0)
rsvm.fit(whas500.x, whas500.y)
pred_nrsvm = nrsvm.predict(kpca.transform(whas500.x))
pred_rsvm = rsvm.predict(whas500.x)
assert len(pred_nrsvm) == len(pred_rsvm)
expected_cindex = concordance_index_censored(whas500.y['fstat'],
whas500.y['lenfol'],
pred_nrsvm)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'],
pred_rsvm, expected_cindex)
def test_survival_squared_hinge_loss(whas500_without_ties):
x, y = whas500_without_ties
nrsvm = NaiveSurvivalSVM(loss='squared_hinge',
dual=False,
tol=8e-7,
max_iter=1000,
random_state=0)
nrsvm.fit(x, y)
rsvm = FastSurvivalSVM(optimizer='avltree',
tol=8e-7,
max_iter=1000,
random_state=0)
rsvm.fit(x, y)
assert_array_almost_equal(nrsvm.coef_.ravel(), rsvm.coef_, 3)
pred_nrsvm = nrsvm.predict(x)
pred_rsvm = rsvm.predict(x)
assert len(pred_nrsvm) == len(pred_rsvm)
expected_cindex = concordance_index_censored(y['fstat'], y['lenfol'],
pred_nrsvm)
assert_cindex_almost_equal(y['fstat'], y['lenfol'], pred_rsvm,
expected_cindex)
def test_compare_builtin_kernel(make_whas500):
whas500 = make_whas500(to_numeric=True)
x = normalize(whas500.x)
rsvm = FastKernelSurvivalSVM(optimizer='rbtree',
kernel="polynomial",
gamma=0.5,
degree=2,
tol=1e-8,
max_iter=100,
random_state=0xf38)
rsvm.fit(x, whas500.y)
pred_rsvm = rsvm.predict(x)
kpca = KernelPCA(kernel="polynomial",
copy_X=True,
gamma=0.5,
degree=2,
random_state=0xf38)
xt = kpca.fit_transform(x)
nrsvm = FastSurvivalSVM(optimizer='rbtree',
tol=1e-8,
max_iter=100,
random_state=0xf38)
nrsvm.fit(xt, whas500.y)
pred_nrsvm = nrsvm.predict(xt)
assert len(pred_nrsvm) == len(pred_rsvm)
expected_cindex = concordance_index_censored(whas500.y['fstat'],
whas500.y['lenfol'],
pred_nrsvm)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'],
pred_rsvm, expected_cindex)
def test_fit_custom_kernel(make_whas500):
whas500 = make_whas500(with_mean=False, with_std=False, to_numeric=True)
alphas = numpy.exp(numpy.linspace(numpy.log(0.001), numpy.log(0.5), 5))
svm_grid = ParameterGrid({"alpha": alphas})
transform = ClinicalKernelTransform(fit_once=True)
transform.prepare(whas500.x_data_frame)
base_estimators = []
for i, params in enumerate(svm_grid):
model = FastSurvivalSVM(max_iter=100, random_state=0, **params)
base_estimators.append(("svm_linear_%d" % i, model))
for i, params in enumerate(svm_grid):
model = FastKernelSurvivalSVM(kernel=transform.pairwise_kernel, max_iter=45, tol=1e-5,
random_state=0, **params)
base_estimators.append(("svm_kernel_%d" % i, model))
cv = KFold(n_splits=3, shuffle=True, random_state=0)
meta = EnsembleSelection(base_estimators, n_estimators=0.4, scorer=score_cindex, cv=cv, n_jobs=4)
meta.fit(whas500.x, whas500.y)
assert len(meta) == 10
assert meta.scores_.shape == (10,)
p = meta.predict(whas500.x)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'], p,
(0.7978084, 59938, 15178, 33, 14))
def test_toy_minlip_predict_1_osqp(self, toy_data):
x, y = toy_data
m = self.minlip_model
m.fit(x, y)
p = m.predict(x)
assert_cindex_almost_equal(y['status'], y['time'], p,
(1.0, 11, 0, 0, 0))
def test_toy_minlip_predict_1_cvxopt(self, toy_data):
x, y = toy_data
m = self.minlip_model
with pytest.deprecated_call():
m.fit(x, y)
p = m.predict(x)
assert_cindex_almost_equal(y['status'], y['time'], p,
(1.0, 11, 0, 0, 0))
def test_breast_cancer_cvxpy(gbsg2):
x, y = gbsg2
m = MinlipSurvivalAnalysis(solver="cvxpy", alpha=1, pairs="next")
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.59576770470121443, 79280, 53792, 0, 32))
def test_breast_cancer_cvxopt(self, gbsg2):
x, y = gbsg2
m = self.model
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.59570007214139709, 79271, 53801, 0, 42))
def test_toy_hinge_fit(self, toy_data):
x, y = toy_data
m = self.svm_model
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['status'], y['time'], p,
(1.0, 11, 0, 0, 0))
def test_fit_no_bootstrap(make_whas500):
whas500 = make_whas500(to_numeric=True)
forest = RandomSurvivalForest(n_estimators=10, bootstrap=False, random_state=2)
forest.fit(whas500.x, whas500.y)
pred = forest.predict(whas500.x)
expected_c = (0.931881994437717, 70030, 5119, 0, 14)
assert_cindex_almost_equal(
whas500.y["fstat"], whas500.y["lenfol"], pred, expected_c)
def test_breast_cancer_ecos(gbsg2):
x, y = gbsg2
x = scale(x)
m = MinlipSurvivalAnalysis(solver="ecos", alpha=1, pairs="next")
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.5990741854033906, 79720, 53352, 0, 42))
def test_fit_spearman_correlation(make_whas500):
whas500 = make_whas500(with_mean=False, with_std=False, to_numeric=True)
meta = _create_survival_ensemble(correlation="spearman")
assert len(meta) == 0
meta.fit(whas500.x, whas500.y)
p = meta.predict(whas500.x)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'], p,
(0.7863312, 59088, 16053, 8, 119))
def test_fit_kendall_correlation(make_whas500):
whas500 = make_whas500(with_mean=False, with_std=False, to_numeric=True)
meta = _create_survival_ensemble(correlation="kendall")
assert len(meta) == 0
meta.fit(whas500.x, whas500.y)
p = meta.predict(whas500.x)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'], p,
(0.7663043, 57570, 17545, 34, 119))
def test_breast_cancer_osqp(gbsg2):
x, y = gbsg2
x = scale(x)
m = MinlipSurvivalAnalysis(solver="osqp", alpha=1, pairs="next")
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.599066670674522, 79719, 53353, 0, 42))
def test_breast_cancer_rbf_cvxpy(gbsg2):
x, y = gbsg2
x = scale(x)
m = MinlipSurvivalAnalysis(solver="cvxpy", alpha=1, kernel="rbf",
gamma=1./8, pairs="next", max_iter=1000)
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.6105867500300589, 81252, 51820, 0, 42))
def test_predict(make_whas500):
whas500 = make_whas500()
model = IPCRidge()
model.fit(whas500.x[:400], whas500.y[:400])
x_test = whas500.x[400:]
y_test = whas500.y[400:]
p = model.predict(x_test)
assert_cindex_almost_equal(y_test['fstat'], y_test['lenfol'], -p,
(0.66925817946226107, 2066, 1021, 0, 1))
assert model.score(x_test, y_test) == 1.0 - 0.66925817946226107
def test_breast_cancer_rbf_cvxopt(self, gbsg2):
x, y = gbsg2
x = scale(x)
m = self.model
m.set_params(kernel="rbf", gamma=1./8)
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.6106092942166647, 81255, 51817, 0, 42))
def test_breast_cancer_rbf_cvxopt(self, gbsg2):
x, y = gbsg2
x = scale(x)
m = self.model
m.set_params(kernel="rbf", gamma=32)
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.6487427858602861, 85974, 46387, 711, 32))
def test_fit(make_whas500):
whas500 = make_whas500(with_mean=False, with_std=False, to_numeric=True)
meta = _create_survival_ensemble()
assert len(meta) == 0
meta.fit(whas500.x, whas500.y)
assert len(meta) == 11
assert meta.scores_.shape == (11,)
p = meta.predict(whas500.x)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'], p,
(0.7863312, 59088, 16053, 8, 119))
def test_predict(make_whas500):
whas500 = make_whas500(with_mean=False, with_std=False, to_numeric=True)
meta = Stacking(MeanEstimator(),
[('coxph', CoxPHSurvivalAnalysis()),
('svm', FastSurvivalSVM(random_state=0))],
probabilities=False)
meta.fit(whas500.x, whas500.y)
# result is different if randomForestSRC has not been compiled with OpenMP support
p = meta.predict(whas500.x)
assert_cindex_almost_equal(whas500.y['fstat'], whas500.y['lenfol'], p,
(0.7848807, 58983, 16166, 0, 14))
def test_fit_predict(make_whas500, name, expected_c):
whas500 = make_whas500(to_numeric=True)
forest = FORESTS[name](random_state=2)
forest.fit(whas500.x, whas500.y)
assert len(forest.estimators_) == 100
pred = forest.predict(whas500.x)
assert numpy.isfinite(pred).all()
assert numpy.all(pred >= 0)
assert_cindex_almost_equal(whas500.y["fstat"], whas500.y["lenfol"], pred,
expected_c)
def test_fit_dropout(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = GradientBoostingSurvivalAnalysis(n_estimators=100, max_features=8,
learning_rate=1.0, dropout_rate=0.03,
random_state=0)
model.fit(whas500_data.x, whas500_data.y)
assert not hasattr(model, "oob_improvement_")
assert model.max_features_ == 8
p = model.predict(whas500_data.x)
assert_cindex_almost_equal(whas500_data.y['fstat'], whas500_data.y['lenfol'], p,
(0.9094333, 68343, 6806, 0, 119))
def test_fit_predict(make_whas500):
whas500 = make_whas500(to_numeric=True)
forest = RandomSurvivalForest(random_state=2)
forest.fit(whas500.x, whas500.y)
assert len(forest.estimators_) == 100
pred = forest.predict(whas500.x)
assert numpy.isfinite(pred).all()
assert numpy.all(pred >= 0)
expected_c = (0.9026201280123488, 67831, 7318, 0, 14)
assert_cindex_almost_equal(whas500.y["fstat"], whas500.y["lenfol"], pred,
expected_c)
def test_toy_hinge_fit(self, toy_data):
x, y = toy_data
m = self.svm_model
sd = numpy.std(x, axis=0)
m.fit(x / sd, y)
assert (1, x.shape[0]) == m.coef_.shape
assert 1 == m.coef0
expected_coef = numpy.array(
[[-1.893832101337, 1.083653895940, 0.810178205398, -2., 2., 0.]])
assert_array_almost_equal(m.coef_, expected_coef)
p = m.predict(x / sd)
assert_cindex_almost_equal(y['status'], y['time'], p,
(1.0, 11, 0, 0, 0))
def test_breast_cancer_rbf_osqp(gbsg2):
x, y = gbsg2
x = scale(x)
m = MinlipSurvivalAnalysis(solver="osqp",
alpha=1,
kernel="rbf",
gamma=1. / 8,
pairs="next",
max_iter=1000)
m.fit(x, y)
assert (1, x.shape[0]) == m.coef_.shape
p = m.predict(x)
assert_cindex_almost_equal(y['cens'], y['time'], p,
(0.6106168089455333, 81256, 51816, 0, 42))
def test_toy_hinge_nearest_fit(self, toy_data):
x, y = toy_data
m = self.svm_model
m.set_params(pairs="nearest")
sd = numpy.std(x, axis=0)
with pytest.deprecated_call():
m.fit(x / sd, y)
assert (1, x.shape[0]) == m.coef_.shape
assert 1 == m.coef0
expected_coef = numpy.array(
[[-1.893832101337, 1.083653895940, 0.810178205398, -2., 2., 0.]])
assert_array_almost_equal(m.coef_, expected_coef, decimal=5)
p = m.predict(x / sd)
assert_cindex_almost_equal(y['status'], y['time'], p,
(1.0, 11, 0, 0, 0))
def test_fit_int_param_as_float(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = GradientBoostingSurvivalAnalysis(n_estimators=100.0, max_depth=3.0, min_samples_split=10.0,
random_state=0)
params = model.get_params()
assert 100 == params["n_estimators"]
assert 3 == params["max_depth"]
assert 10 == params["min_samples_split"]
model.set_params(max_leaf_nodes=15.0)
assert 15 == model.get_params()["max_leaf_nodes"]
model.fit(whas500_data.x, whas500_data.y)
p = model.predict(whas500_data.x)
assert_cindex_almost_equal(whas500_data.y['fstat'], whas500_data.y['lenfol'], p,
(0.90256690042449006, 67826, 7321, 2, 119))
