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_compare_clinical_kernel(self):
x_full, y = load_whas500()
trans = ClinicalKernelTransform()
trans.fit(x_full)
kpca = KernelPCA(kernel=trans.pairwise_kernel, copy_X=True)
xt = kpca.fit_transform(self.x)
nrsvm = FastSurvivalSVM(optimizer='rbtree',
tol=1e-8,
max_iter=500,
random_state=0)
nrsvm.fit(xt, y)
rsvm = FastKernelSurvivalSVM(optimizer='rbtree',
kernel=trans.pairwise_kernel,
tol=1e-8,
max_iter=500,
random_state=0)
rsvm.fit(self.x.values, y)
pred_nrsvm = nrsvm.predict(kpca.transform(self.x))
pred_rsvm = rsvm.predict(self.x.values)
self.assertEqual(len(pred_nrsvm), len(pred_rsvm))
c1 = concordance_index_censored(y['fstat'], y['lenfol'], pred_nrsvm)
c2 = concordance_index_censored(y['fstat'], y['lenfol'], pred_rsvm)
self.assertAlmostEqual(c1[0], c2[0])
self.assertTupleEqual(c1[1:], c2[1:])
def test_survival_squared_hinge_loss(self):
x, y = self.get_data_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)
self.assertEqual(len(pred_nrsvm), len(pred_rsvm))
c1 = concordance_index_censored(y['fstat'], y['lenfol'], pred_nrsvm)
c2 = concordance_index_censored(y['fstat'], y['lenfol'], pred_rsvm)
self.assertAlmostEqual(c1[0], c2[0])
self.assertTupleEqual(c1[1:], c2[1:])
def test_fit_and_predict_hybrid(self):
if self.OPTIMIZER in {'simple', 'PRSVM'}:
raise unittest.SkipTest("regression not implemented for " +
self.OPTIMIZER)
ssvm = FastSurvivalSVM(optimizer=self.OPTIMIZER,
rank_ratio=0.5,
max_iter=50,
fit_intercept=True,
random_state=0)
ssvm.fit(self.x.values, self.y)
self.assertAlmostEqual(6.1409367385513729, ssvm.intercept_)
expected_coef = numpy.array([
-0.0209254120718, -0.265768317208, -0.154254689136,
0.0800600947891, -0.290121131022, -0.0288851785213,
0.0998004550073, 0.0454100937492, -0.125863947621, 0.0343588337797,
-0.000710219364914, 0.0546969104996, -0.5375338235,
-0.0137995110308
])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(self.x.values)
rmse = numpy.sqrt(mean_squared_error(self.y['lenfol'], pred))
self.assertAlmostEqual(780.52617631863893, rmse)
def test_compare_builtin_kernel(self):
x = normalize(self.x)
y = self.y
rsvm = FastKernelSurvivalSVM(optimizer='rbtree',
kernel="polynomial",
gamma=0.5,
degree=2,
tol=1e-8,
max_iter=100,
random_state=0xf38)
rsvm.fit(x, 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, y)
pred_nrsvm = nrsvm.predict(xt)
self.assertEqual(len(pred_nrsvm), len(pred_rsvm))
c1 = concordance_index_censored(y['fstat'], y['lenfol'], pred_nrsvm)
c2 = concordance_index_censored(y['fstat'], y['lenfol'], pred_rsvm)
self.assertAlmostEqual(c1[0], c2[0])
self.assertTupleEqual(c1[1:], c2[1:])
def test_default_optimizer(make_whas500):
whas500 = make_whas500(to_numeric=True)
ssvm = FastSurvivalSVM(tol=1e-4, max_iter=25)
with warnings.catch_warnings():
warnings.simplefilter("ignore", category=ConvergenceWarning)
ssvm.fit(whas500.x, whas500.y)
assert 'avltree' == ssvm.optimizer
def test_unknown_optimizer(fake_data):
x, y = fake_data
ssvm = FastSurvivalSVM(rank_ratio=0, optimizer='random stuff')
with pytest.raises(ValueError,
match="unknown optimizer: random stuff"):
ssvm.fit(x, y)
def test_regression_not_supported(fake_data, value):
x, y = fake_data
ssvm = FastSurvivalSVM(rank_ratio=0, optimizer=value)
with pytest.raises(ValueError,
match="optimizer {!r} does not implement regression objective".format(value)):
ssvm.fit(x, y)
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_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_rank_ratio_out_of_bounds(fake_data, value):
x, y = fake_data
ssvm = FastSurvivalSVM(rank_ratio=value)
with pytest.raises(ValueError,
match=r"rank_ratio must be in \[0; 1\]"):
ssvm.fit(x, y)
def test_compare_rbf(self):
x, y = load_whas500()
x = encode_categorical(standardize(x))
kpca = KernelPCA(kernel="rbf")
xt = kpca.fit_transform(x)
nrsvm = FastSurvivalSVM(optimizer='rbtree',
tol=1e-8,
max_iter=1000,
random_state=0)
nrsvm.fit(xt, y)
rsvm = FastKernelSurvivalSVM(optimizer='rbtree',
kernel="rbf",
tol=1e-8,
max_iter=1000,
random_state=0)
rsvm.fit(x, y)
pred_nrsvm = nrsvm.predict(kpca.transform(x))
pred_rsvm = rsvm.predict(x)
self.assertEqual(len(pred_nrsvm), len(pred_rsvm))
c1 = concordance_index_censored(y['fstat'], y['lenfol'], pred_nrsvm)
c2 = concordance_index_censored(y['fstat'], y['lenfol'], pred_rsvm)
self.assertAlmostEqual(c1[0], c2[0])
self.assertTupleEqual(c1[1:], c2[1:])
def test_negative_time():
x = numpy.arange(80).reshape(10, 8)
y = Surv.from_arrays([0, 1, 0, 1, 1, 0, 1, 0, 0, 1], [1, 1, -2, 1, 1, 6, 1, 2, 3, 1])
rsvm = FastSurvivalSVM(rank_ratio=0.5)
with pytest.raises(ValueError,
match="observed time contains values smaller or equal to zero"):
rsvm.fit(x, y)
def test_fit_timeit(self):
rnd = numpy.random.RandomState(0)
idx = rnd.choice(numpy.arange(self.x.shape[0]), replace=False, size=100)
ssvm = FastSurvivalSVM(optimizer=self.OPTIMIZER, timeit=3, random_state=0)
ssvm.fit(self.x.values[idx, :], self.y[idx])
self.assertTrue('timings' in ssvm.optimizer_result_)
def test_ranking_with_fit_intercept():
x = numpy.zeros((100, 10))
y = Surv.from_arrays(numpy.ones(100, dtype=bool), numpy.arange(1, 101, dtype=float))
ssvm = FastSurvivalSVM(rank_ratio=1.0, fit_intercept=True)
with pytest.raises(ValueError,
match="fit_intercept=True is only meaningful if rank_ratio < 1.0"):
ssvm.fit(x, y)
def test_all_censored():
x = numpy.arange(80).reshape(10, 8)
y = Surv.from_arrays(numpy.zeros(10, dtype=bool), [0, 1, 2, 1, 1, 0, 1, 2, 3, 1])
rsvm = FastSurvivalSVM()
with pytest.raises(ValueError,
match="all samples are censored"):
rsvm.fit(x, y)
def test_y_invalid(y):
x = numpy.zeros((100, 10))
rsvm = FastSurvivalSVM()
with pytest.raises(ValueError,
match='y must be a structured array with the first field'
' being a binary class event indicator and the second field'
' the time of the event/censoring'):
rsvm.fit(x, y)
def test_fit_timeit(make_whas500, optimizer_any):
whas500 = make_whas500(to_numeric=True)
rnd = numpy.random.RandomState(0)
idx = rnd.choice(numpy.arange(whas500.x.shape[0]), replace=False, size=100)
ssvm = FastSurvivalSVM(optimizer=optimizer_any, timeit=3, random_state=0)
ssvm.fit(whas500.x[idx, :], whas500.y[idx])
assert 'timings' in ssvm.optimizer_result_
def test_event_not_binary():
x = numpy.arange(80).reshape(10, 8)
y = numpy.empty(dtype=[('event', int), ('time', float)], shape=10)
y['event'] = numpy.array([0, 1, 2, 1, 1, 0, 1, 2, 3, 1], dtype=int)
y['time'] = numpy.arange(10)
rsvm = FastSurvivalSVM()
with pytest.raises(ValueError,
match="elements of event indicator must be boolean, but found int"):
rsvm.fit(x, y)
def test_time_not_numeric():
x = numpy.arange(80).reshape(10, 8)
y = numpy.empty(dtype=[('event', bool), ('time', bool)], shape=10)
y['event'] = numpy.array([0, 1, 0, 1, 1, 0, 1, 0, 0, 1], dtype=bool)
y['time'] = numpy.ones(10, dtype=bool)
rsvm = FastSurvivalSVM()
with pytest.raises(ValueError,
match="time must be numeric, but found bool"):
rsvm.fit(x, y)
def test_fit_and_predict_ranking(self):
ssvm = FastSurvivalSVM(optimizer=self.OPTIMIZER, random_state=0)
ssvm.fit(self.x.values, self.y)
self.assertFalse(hasattr(ssvm, "intercept_"))
expected_coef = numpy.array([-0.02066177, -0.26449933, -0.15205399, 0.0794547, -0.28840498, -0.02864288,
0.09901995, 0.04505302, -0.12512215, 0.03341365, -0.00110442, 0.05446756,
-0.53009875, -0.01394175])
assert_array_almost_equal(expected_coef, ssvm.coef_)
self.assertEquals(self.x.shape[1], ssvm.coef_.shape[0])
c = ssvm.score(self.x.values, self.y)
self.assertAlmostEqual(0.7860650174985695, c, 6)
def test_fit_and_predict_regression_no_intercept(make_whas500, optimizer_regression):
whas500 = make_whas500(to_numeric=True)
ssvm = FastSurvivalSVM(optimizer=optimizer_regression, rank_ratio=0.0,
max_iter=50, fit_intercept=False, random_state=0)
ssvm.fit(whas500.x, whas500.y)
assert not hasattr(ssvm, "intercept_")
expected_coef = numpy.array([1.39989875, -1.16903161, -0.40195857, -0.05848903, -0.08421557, 4.11924729,
0.25135451, 1.89067276, -0.25751401, -0.10213143, 1.56333622, 3.10136873,
-2.23644848, -0.11620715])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(whas500.x)
rmse = numpy.sqrt(mean_squared_error(whas500.y['lenfol'], pred))
assert round(abs(15838.510668936022 - rmse), 7) == 0
def test_fit_and_predict_hybrid_no_intercept(make_whas500, optimizer_regression):
whas500 = make_whas500(to_numeric=True)
ssvm = FastSurvivalSVM(optimizer=optimizer_regression, rank_ratio=0.5,
max_iter=50, fit_intercept=False, random_state=0)
ssvm.fit(whas500.x, whas500.y)
assert not hasattr(ssvm, "intercept_")
expected_coef = numpy.array([0.00669121, -0.2754864, -0.14124808, 0.0748376, -0.2812598, 0.07543884,
0.09845683, 0.08398258, -0.12182314, 0.02637739, 0.03060149, 0.11870598,
-0.52688224, -0.01762842])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(whas500.x)
rmse = numpy.sqrt(mean_squared_error(whas500.y['lenfol'], pred))
assert round(abs(1128.4460587629746 - rmse), 7) == 0
def test_fit_and_predict_ranking(make_whas500, optimizer_any):
whas500 = make_whas500(to_numeric=True)
ssvm = FastSurvivalSVM(optimizer=optimizer_any, random_state=0)
ssvm.fit(whas500.x, whas500.y)
assert not hasattr(ssvm, "intercept_")
expected_coef = numpy.array([-0.02066177, -0.26449933, -0.15205399, 0.0794547, -0.28840498, -0.02864288,
0.09901995, 0.04505302, -0.12512215, 0.03341365, -0.00110442, 0.05446756,
-0.53009875, -0.01394175])
assert_array_almost_equal(expected_coef, ssvm.coef_)
assert whas500.x.shape[1] == ssvm.coef_.shape[0]
c = ssvm.score(whas500.x, whas500.y)
assert round(abs(0.7860650174985695 - c), 6) == 0
def test_fit_and_predict_hybrid_no_intercept(self):
if self.OPTIMIZER in {'simple', 'PRSVM'}:
raise unittest.SkipTest("regression not implemented for " + self.OPTIMIZER)
ssvm = FastSurvivalSVM(optimizer=self.OPTIMIZER, rank_ratio=0.5,
max_iter=50, fit_intercept=False, random_state=0)
ssvm.fit(self.x.values, self.y)
self.assertFalse(hasattr(ssvm, "intercept_"))
expected_coef = numpy.array([0.00669121, -0.2754864, -0.14124808, 0.0748376, -0.2812598, 0.07543884,
0.09845683, 0.08398258, -0.12182314, 0.02637739, 0.03060149, 0.11870598,
-0.52688224, -0.01762842])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(self.x.values)
rmse = numpy.sqrt(mean_squared_error(self.y['lenfol'], pred))
self.assertAlmostEqual(1128.4460587629746, rmse)
def test_fit_and_predict_regression_no_intercept(self):
if self.OPTIMIZER in {'simple', 'PRSVM'}:
raise unittest.SkipTest("regression not implemented for " + self.OPTIMIZER)
ssvm = FastSurvivalSVM(optimizer=self.OPTIMIZER, rank_ratio=0.0,
max_iter=50, fit_intercept=False, random_state=0)
ssvm.fit(self.x.values, self.y)
self.assertFalse(hasattr(ssvm, "intercept_"))
expected_coef = numpy.array([1.39989875, -1.16903161, -0.40195857, -0.05848903, -0.08421557, 4.11924729,
0.25135451, 1.89067276, -0.25751401, -0.10213143, 1.56333622, 3.10136873,
-2.23644848, -0.11620715])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(self.x.values)
rmse = numpy.sqrt(mean_squared_error(self.y['lenfol'], pred))
self.assertAlmostEqual(15838.510668936022, rmse)
def test_fit_and_predict_regression(make_whas500, optimizer_regression):
whas500 = make_whas500(to_numeric=True)
ssvm = FastSurvivalSVM(optimizer=optimizer_regression, rank_ratio=0.0,
max_iter=50, fit_intercept=True, random_state=0)
ssvm.fit(whas500.x, whas500.y)
assert round(abs(6.4160179606675278 - ssvm.intercept_), 7) == 0
expected_coef = numpy.array(
[-0.0730891368237, -0.536630355029, -0.497411603275, 0.269039958377, -0.730559850692, -0.0148443526234,
0.285916578892, 0.165960302339, -0.301749910087, 0.334855938531, 0.0886214732161, 0.0554890272028,
-2.12680470014, 0.0421466831393
])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(whas500.x)
rmse = numpy.sqrt(mean_squared_error(whas500.y['lenfol'], pred))
assert round(abs(1206.6556186869332 - rmse), 7) == 0
def test_fit_and_predict_hybrid(make_whas500, optimizer_regression):
whas500 = make_whas500(to_numeric=True)
ssvm = FastSurvivalSVM(optimizer=optimizer_regression, rank_ratio=0.5,
max_iter=50, fit_intercept=True, random_state=0)
ssvm.fit(whas500.x, whas500.y)
assert round(abs(6.1409367385513729 - ssvm.intercept_), 7) == 0
expected_coef = numpy.array(
[-0.0209254120718, -0.265768317208, -0.154254689136, 0.0800600947891, -0.290121131022, -0.0288851785213,
0.0998004550073, 0.0454100937492, -0.125863947621, 0.0343588337797, -0.000710219364914, 0.0546969104996,
-0.5375338235, -0.0137995110308
])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(whas500.x)
rmse = numpy.sqrt(mean_squared_error(whas500.y['lenfol'], pred))
assert round(abs(780.52617631863893 - rmse), 7) == 0
def test_fit_and_predict_regression(self):
if self.OPTIMIZER in {'simple', 'PRSVM'}:
raise unittest.SkipTest("regression not implemented for " + self.OPTIMIZER)
ssvm = FastSurvivalSVM(optimizer=self.OPTIMIZER, rank_ratio=0.0,
max_iter=50, fit_intercept=True, random_state=0)
ssvm.fit(self.x.values, self.y)
self.assertAlmostEqual(6.4160179606675278, ssvm.intercept_)
expected_coef = numpy.array(
[-0.0730891368237, -0.536630355029, -0.497411603275, 0.269039958377, -0.730559850692, -0.0148443526234,
0.285916578892, 0.165960302339, -0.301749910087, 0.334855938531, 0.0886214732161, 0.0554890272028,
-2.12680470014, 0.0421466831393
])
assert_array_almost_equal(expected_coef, ssvm.coef_)
pred = ssvm.predict(self.x.values)
rmse = numpy.sqrt(mean_squared_error(self.y['lenfol'], pred))
self.assertAlmostEqual(1206.6556186869332, rmse)
def test_compare_rbf(self):
x = normalize(self.x)
y = self.y
rsvm = FastKernelSurvivalSVM(optimizer='rbtree', kernel="rbf",
tol=1e-6, max_iter=65, random_state=0)
rsvm.fit(x, y)
kpca = KernelPCA(kernel="rbf", copy_X=True)
xt = kpca.fit_transform(x)
nrsvm = FastSurvivalSVM(optimizer='rbtree', tol=1e-6, max_iter=30, random_state=0)
nrsvm.fit(xt, y)
pred_nrsvm = nrsvm.predict(xt)
pred_rsvm = rsvm.predict(x)
self.assertEqual(len(pred_nrsvm), len(pred_rsvm))
c1 = concordance_index_censored(y['fstat'], y['lenfol'], pred_nrsvm)
c2 = concordance_index_censored(y['fstat'], y['lenfol'], pred_rsvm)
self.assertAlmostEqual(c1[0], c2[0])
self.assertTupleEqual(c1[1:], c2[1:])
