def test_compare_rbf(self):
x, y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'], '1')
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 main(args):
LOG.info("Using IPython profile %s", args.profile)
rc = parallel.Client(profile=args.profile)
with rc[:].sync_imports():
from sklearn.metrics import mean_squared_error
import numpy
_x, _y, _x_test, _y_test = load_arff_file(args.input,
[args.event, args.time],
args.outcome,
args.test,
to_numeric=False)
_data = _x.copy()
_x = categorical_to_numeric(_x)
if _x_test is not None:
_x_test = categorical_to_numeric(_x_test)
_estimator = get_estimator(args.method, args.seed, _data)
_param_grid = get_param_grid(args.params)
print_settings(_estimator, _x, _y)
_grid_search = run_grid_search(_estimator, _param_grid, _x, _y, _x_test,
_y_test, args.seed, args.profile)
if args.test is None:
_output = "results-rmse-%s-%s.csv" % (basename(
args.input).rstrip(".arff"), args.method)
else:
_output = "results-rmse-%s+%s-%s.csv" % (basename(
args.input).rstrip(".arff"), basename(
args.test).rstrip(".arff"), args.method)
write_results(_grid_search.best_params_, _output)
rc[:].clear()
def test_compare_clinical_kernel(self):
x_full, y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'], '1',
standardize_numeric=False, to_numeric=False)
trans = ClinicalKernelTransform()
trans.fit(x_full)
x = encode_categorical(standardize(x_full))
kpca = KernelPCA(kernel=trans.pairwise_kernel)
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=trans.pairwise_kernel,
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 main(args):
_x, _y, _x_test, _y_test = load_arff_file(
args.input, [args.event, 'ENDTRS_C', 'ENTRT_PC'],
survival=False,
path_testing=args.test,
to_numeric=False)
model = fit_and_dump(_x, _y, args)
model.load_from_db()
print("Ensemble size: %d" % len(model))
if _x_test is not None:
_x_test = categorical_to_numeric(_x_test)
proba = model.predict(_x_test.values)
pred_labels = model.classes_.take(numpy.argmax(proba, axis=1), axis=0)
result = pandas.DataFrame({
"RISK": proba[:, model.classes_ == 1].ravel(),
"RPT": _x_test.index.to_series(),
"DISCONT": pred_labels
})
result.set_index("RPT", inplace=True)
_results_file = 'results-%s-%s.csv' % (basename(
args.input), "ensemble_selection_classification")
result.to_csv(_results_file)
def test_load_with_categorical_index_2(self):
fp = StringIO(ARFF_CATEGORICAL_INDEX_2)
x_train, y_train, x_test, y_test = load_arff_file(fp, ["label"], pos_label="yes", survival=False,
standardize_numeric=False, to_numeric=False)
self.assertEqual(x_test, None)
self.assertEqual(y_test, None)
self.assertTupleEqual(x_train.shape, (5, 2))
self.assertTupleEqual(y_train.shape, (5, 1))
index = pandas.Index(['ASampleOne', 'ASampleTwo', 'ASampleThree', 'ASampleFour', 'ASampleFive'],
name='index', dtype=object)
tm.assert_index_equal(x_train.index, index, exact=True)
label = pandas.Series(pandas.Categorical(["no", "no", "yes", "yes", "no"], categories=["yes", "no"], ordered=False),
name="label", index=index)
tm.assert_series_equal(y_train["label"], label, check_exact=True)
value = pandas.Series([1.51, 1.38, -20, 245.3, 3.14], name="value", index=index)
tm.assert_series_equal(x_train["value"], value, check_exact=True)
size = pandas.Series(pandas.Categorical(["small", "small", "large", "small", "large"],
categories=["small", "medium", "large"], ordered=False),
name="size", index=index)
tm.assert_series_equal(x_train["size"], size, check_exact=True)
def setUp(self):
# naive survival SVM does resolve ties in survival time differently,
# therefore use data without ties
self.x, self.y, _, _ = load_arff_file(WHAS500_NOTIES_FILE,
['fstat', 'lenfol'],
'1',
standardize_numeric=False)
def main(args):
LOG.info("Using IPython profile %s", args.profile)
rc = parallel.Client(profile=args.profile)
_x, _y, _x_test, _y_test = load_arff_file(
args.input, [args.event, 'ENDTRS_C', 'ENTRT_PC'],
survival=False,
path_testing=args.test,
to_numeric=False)
_data = _x.copy()
_x = categorical_to_numeric(_x)
if _x_test is not None:
_x_test = categorical_to_numeric(_x_test)
_y = _y[args.event].cat.codes.values
_estimator = get_estimator(args.method, args.seed, _data)
_param_grid = get_param_grid(args.params)
print_settings(_estimator, _x, _y)
_grid_search = run_grid_search(_estimator, _param_grid, args.metric, _x,
_y, _x_test, _y_test, args.seed,
args.profile)
if args.test is None:
_output = "results-%s-%s-%s.csv" % (args.metric, basename(
args.input).rstrip(".arff"), args.method)
else:
_output = "results-%s-%s+%s-%s.csv" % (args.metric, basename(
args.input).rstrip(".arff"), basename(
args.test).rstrip(".arff"), args.method)
write_results(_grid_search.best_params_, _output)
rc[:].clear()
def test_load_train_and_test_with_different_columns(self):
tmp_train = tempfile.NamedTemporaryFile("w", suffix=".arff", delete=False)
tmp_test = tempfile.NamedTemporaryFile("w", suffix=".arff", delete=False)
try:
_make_and_write_data(tmp_train, 100, 19, False, True, 0)
_make_and_write_data(tmp_test, 20, 11, False, True, 0)
with warnings.catch_warnings(record=True) as w:
warnings.simplefilter("always")
load_arff_file(tmp_train.name, ["event", "time"], 1,
path_testing=tmp_test.name,
survival=True,
standardize_numeric=False, to_numeric=False)
self.assertEqual(1, len(w))
self.assertEqual("Restricting columns to intersection between training and testing data",
str(w[0].message))
finally:
os.unlink(tmp_train.name)
os.unlink(tmp_test.name)
def test_load_train_and_test_with_categorical_index(self):
fp_1 = StringIO(ARFF_CATEGORICAL_INDEX_1)
fp_2 = StringIO(ARFF_CATEGORICAL_INDEX_2)
x_train, y_train, x_test, y_test = load_arff_file(fp_1, ["label"], pos_label="yes",
path_testing=fp_2, survival=False,
standardize_numeric=False, to_numeric=False)
self.assertTupleEqual(x_train.shape, (4, 2))
self.assertTupleEqual(x_test.shape, (5, 2))
self.assertTupleEqual(y_train.shape, (4, 1))
self.assertTupleEqual(y_test.shape, (5, 1))
# Check train data
train_index = pandas.Index(['SampleOne', 'SampleTwo', 'SampleThree', 'SampleFour'],
name='index', dtype=object)
tm.assert_index_equal(x_train.index, train_index, exact=True)
train_label = pandas.Series(
pandas.Categorical(["yes", "no", "yes", "yes"], categories=["no", "yes"], ordered=False),
name="label", index=train_index)
tm.assert_series_equal(y_train["label"], train_label, check_exact=True)
train_value = pandas.Series([15.1, 13.8, -0.2, 2.453], name="value", index=train_index)
tm.assert_series_equal(x_train["value"], train_value, check_exact=True)
train_size = pandas.Series(pandas.Categorical(["medium", "large", "small", "large"],
categories=["small", "medium", "large"], ordered=False),
name="size", index=train_index)
tm.assert_series_equal(x_train["size"], train_size, check_exact=True)
# Check test data
test_index = pandas.Index(['ASampleOne', 'ASampleTwo', 'ASampleThree', 'ASampleFour', 'ASampleFive'],
name='index', dtype=object)
tm.assert_index_equal(x_test.index, test_index, exact=True)
test_label = pandas.Series(
pandas.Categorical(["no", "no", "yes", "yes", "no"], categories=["yes", "no"], ordered=False),
name="label", index=test_index)
tm.assert_series_equal(y_test["label"], test_label, check_exact=True)
test_value = pandas.Series([1.51, 1.38, -20, 245.3, 3.14], name="value", index=test_index)
tm.assert_series_equal(x_test["value"], test_value, check_exact=True)
test_size = pandas.Series(pandas.Categorical(["small", "small", "large", "small", "large"],
categories=["small", "medium", "large"], ordered=False),
name="size", index=test_index)
tm.assert_series_equal(x_test["size"], test_size, check_exact=True)
def main(args):
_x, _y, _x_test, _y_test = load_arff_file(args.input, [args.event, args.time], args.outcome,
args.test, to_numeric=False)
model = fit_and_dump(_x, _y, args)
print("Ensemble size: %d" % len(model))
if _x_test is not None:
_x_test = categorical_to_numeric(_x_test)
p = numpy.exp(model.predict(_x_test.values))
result = pandas.DataFrame({"TIMETOEVENT": p, "RPT": _x_test.index.to_series()})
result.set_index("RPT", inplace=True)
_results_file = 'results-%s-%s.csv' % (basename(args.input), "ensemble_selection_regression")
result.to_csv(_results_file)
def test_fit_and_predict_clinical_kernel(self):
x_full, y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'], '1',
standardize_numeric=False, to_numeric=False)
trans = ClinicalKernelTransform()
trans.fit(x_full)
x = encode_categorical(standardize(x_full))
ssvm = FastKernelSurvivalSVM(optimizer="rbtree", kernel=trans.pairwise_kernel, max_iter=100, random_state=0)
ssvm.fit(x.values, y)
self.assertFalse(ssvm._pairwise)
self.assertEquals(x.shape[0], ssvm.coef_.shape[0])
c = ssvm.score(x.values, y)
self.assertLessEqual(abs(0.83699051218246412 - c), 1e-3)
def test_load_with_index(self):
tmp = tempfile.NamedTemporaryFile("w", suffix=".arff", delete=False)
try:
dataset = _make_and_write_data(tmp, 100, 10, True, True, 0)
x_train, y_train, x_test, y_test = load_arff_file(tmp.name, ["event", "time"], 1, survival=True,
standardize_numeric=False, to_numeric=False)
self.assertEqual(x_test, None)
self.assertEqual(y_test, None)
cols = ["event", "time"]
x_true = dataset.drop(cols, axis=1)
self.assert_x_equal(x_true, x_train)
self.assert_y_equal(dataset, y_train)
finally:
os.unlink(tmp.name)
def main(args):
LOG.info("Using IPython profile %s", args.profile)
rc = parallel.Client(profile=args.profile)
with rc[:].sync_imports():
from survival.metrics import concordance_index_censored
import numpy
if args.metric == "timeroc":
with rc[:].sync_imports():
from rpy2.robjects.packages import importr
from rpy2.robjects import numpy2ri
from scipy.integrate import simps
_x, _y, _x_test, _y_test = load_arff_file(args.input,
[args.event, args.time],
args.outcome,
args.test,
to_numeric=False)
_data = _x.copy()
_x = categorical_to_numeric(_x)
if _x_test is not None:
_x_test = categorical_to_numeric(_x_test)
_estimator = get_estimator(args.method, args.seed, _data)
_param_grid = get_param_grid(args.params)
print_settings(_estimator, _x, _y)
_grid_search = run_grid_search(_estimator, _param_grid, args.metric, _x,
_y, _x_test, _y_test, args.seed,
args.profile)
if args.test is None:
_output = "results-%s-%s-%s.csv" % (args.metric, basename(
args.input).rstrip(".arff"), args.method)
else:
_output = "results-%s-%s+%s-%s.csv" % (args.metric, basename(
args.input).rstrip(".arff"), basename(
args.test).rstrip(".arff"), args.method)
write_results(_grid_search.best_params_, _output)
rc[:].clear()
def test_fit_and_predict_clinical_kernel(self):
x_full, y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'],
'1',
standardize_numeric=False,
to_numeric=False)
trans = ClinicalKernelTransform()
trans.fit(x_full)
x = encode_categorical(standardize(x_full))
ssvm = FastKernelSurvivalSVM(optimizer="rbtree",
kernel=trans.pairwise_kernel,
max_iter=100,
random_state=0)
ssvm.fit(x.values, y)
self.assertFalse(ssvm._pairwise)
self.assertEquals(x.shape[0], ssvm.coef_.shape[0])
c = ssvm.score(x.values, y)
self.assertLessEqual(abs(0.83699051218246412 - c), 1e-3)
def test_load_train_and_test_no_labels(self):
tmp_train = tempfile.NamedTemporaryFile("w", suffix=".arff", delete=False)
tmp_test = tempfile.NamedTemporaryFile("w", suffix=".arff", delete=False)
try:
train_dataset = _make_and_write_data(tmp_train, 100, 10, True, True, 0)
test_dataset = _make_and_write_data(tmp_test, 20, 10, True, False, 0)
x_train, y_train, x_test, y_test = load_arff_file(tmp_train.name, ["event", "time"], 1,
path_testing=tmp_test.name,
survival=True,
standardize_numeric=False, to_numeric=False)
cols = ["event", "time"]
x_true = train_dataset.drop(cols, axis=1)
self.assert_x_equal(x_true, x_train)
self.assert_y_equal(train_dataset, y_train)
self.assert_x_equal(test_dataset, x_test)
self.assertEqual(y_test, None)
finally:
os.unlink(tmp_train.name)
os.unlink(tmp_test.name)
def test_compare_rbf(self):
x, y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'], '1')
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_compare_clinical_kernel(self):
x_full, y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'],
'1',
standardize_numeric=False,
to_numeric=False)
trans = ClinicalKernelTransform()
trans.fit(x_full)
x = encode_categorical(standardize(x_full))
kpca = KernelPCA(kernel=trans.pairwise_kernel)
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=trans.pairwise_kernel,
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 setUp(self):
# naive survival SVM does resolve ties in survival time differently,
# therefore use data without ties
self.x, self.y, _, _ = load_arff_file(WHAS500_NOTIES_FILE, ['fstat', 'lenfol'], '1',
standardize_numeric=False)
def setUp(self):
x, y, _, _ = load_arff_file(GBSG2_FILE,
["cens", "time"], "1", standardize_numeric=False, to_numeric=False)
self.x = encode_categorical(x)
self.y = y
def setUp(self):
self.x, self.y, _, _ = load_arff_file(WHAS500_FILE, ['fstat', 'lenfol'], '1',
standardize_numeric=True)
def setUp(self):
self.x, self.y, _, _ = load_arff_file(WHAS500_FILE,
['fstat', 'lenfol'],
'1',
standardize_numeric=True)