scikit-survival 0.26.0__tar.gz → 0.27.0__tar.gz

{scikit_survival-0.26.0/scikit_survival.egg-info → scikit_survival-0.27.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scikit-survival
-Version: 0.26.0
+Version: 0.27.0
 Summary: Survival analysis built on top of scikit-learn
 Author-email: Sebastian Pölsterl <sebp@k-d-w.org>
 License-Expression: GPL-3.0-or-later
@@ -31,16 +31,16 @@ License-File: COPYING
 Requires-Dist: ecos
 Requires-Dist: joblib
 Requires-Dist: numexpr
-Requires-Dist: numpy
+Requires-Dist: numpy>=2.0.0
 Requires-Dist: osqp>=1.0.2
-Requires-Dist: pandas>=2.0.0
-Requires-Dist: scipy>=1.3.2
-Requires-Dist: scikit-learn<1.8,>=1.6.1
+Requires-Dist: pandas>=2.2.0
+Requires-Dist: scipy>=1.13.0
+Requires-Dist: scikit-learn<1.9,>=1.8.0
 Dynamic: license-file
 
 |License| |Docs| |DOI|
 
-|build-tests| |build-windows| |Codecov| |Codacy|
+|build-tests| |Codecov| |Codacy|
 
 ***************
 scikit-survival
@@ -76,10 +76,10 @@ Requirements
 - ecos
 - joblib
 - numexpr
-- numpy
+- numpy 2.0.0 or later
 - osqp
-- pandas 2.0.0 or later
-- scikit-learn 1.6 or 1.7
+- pandas 2.2.0 or later
+- scikit-learn 1.8
 - scipy
 - C/C++ compiler
 
@@ -178,8 +178,4 @@ Please cite the following paper if you are using **scikit-survival**.
   :target: https://github.com/sebp/scikit-survival/actions?query=workflow%3Atests+branch%3Amaster
   :alt: GitHub Actions Tests Status
 
-.. |build-windows| image:: https://ci.appveyor.com/api/projects/status/github/sebp/scikit-survival?branch=master&svg=true
-   :target: https://ci.appveyor.com/project/sebp/scikit-survival
-   :alt: Windows Build Status on AppVeyor
-
 .. _survival analysis: https://en.wikipedia.org/wiki/Survival_analysis

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/README.rst

@@ -1,6 +1,6 @@
 |License| |Docs| |DOI|
 
-|build-tests| |build-windows| |Codecov| |Codacy|
+|build-tests| |Codecov| |Codacy|
 
 ***************
 scikit-survival
@@ -36,10 +36,10 @@ Requirements
 - ecos
 - joblib
 - numexpr
-- numpy
+- numpy 2.0.0 or later
 - osqp
-- pandas 2.0.0 or later
-- scikit-learn 1.6 or 1.7
+- pandas 2.2.0 or later
+- scikit-learn 1.8
 - scipy
 - C/C++ compiler
 
@@ -138,8 +138,4 @@ Please cite the following paper if you are using **scikit-survival**.
   :target: https://github.com/sebp/scikit-survival/actions?query=workflow%3Atests+branch%3Amaster
   :alt: GitHub Actions Tests Status
 
-.. |build-windows| image:: https://ci.appveyor.com/api/projects/status/github/sebp/scikit-survival?branch=master&svg=true
-   :target: https://ci.appveyor.com/project/sebp/scikit-survival
-   :alt: Windows Build Status on AppVeyor
-
 .. _survival analysis: https://en.wikipedia.org/wiki/Survival_analysis

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/doc/conf.py

@@ -217,7 +217,7 @@ extlinks = {
 }
 
 intersphinx_mapping = {
-    "sklearn": ("https://scikit-learn.org/1.7", None),
+    "sklearn": ("https://scikit-learn.org/1.8", None),
     "cython": ("https://cython.readthedocs.io/en/latest/", None),
     "numpy": ("https://numpy.org/doc/stable/", None),
     "scipy": ("https://docs.scipy.org/doc/scipy/", None),

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/doc/install.rst

@@ -97,9 +97,9 @@ The current minimum dependencies to run scikit-survival are:
 - ecos
 - joblib
 - numexpr
-- numpy
+- numpy 2.0.0 or later
 - osqp
-- pandas 2.0.0 or later
-- scikit-learn 1.6 or 1.7
+- pandas 2.2.0 or later
+- scikit-learn 1.8
 - scipy
 - C/C++ compiler

scikit_survival-0.27.0/doc/release_notes/v0.27.rst

@@ -0,0 +1,47 @@
+.. _release_notes_0_27:
+
+What's new in 0.27
+==================
+
+scikit-survival 0.27.0 (2026-02-02)
+-----------------------------------
+
+This release adds support for scikit-learn 1.8 and pandas 3.0.
+The updated minimum supported version of dependencies are:
+
+   +--------------+-----------------+
+   | Package      | Minimum Version |
+   +==============+=================+
+   +--------------+-----------------+
+   | scikit-learn | 1.8.0           |
+   +--------------+-----------------+
+   | pandas       | 2.2.0           |
+   +--------------+-----------------+
+   | scipy        | 1.13.0          |
+   +--------------+-----------------+
+   | numpy        | 2.0.0           |
+   +--------------+-----------------+
+
+Enhancements
+^^^^^^^^^^^^
+
+- Add support for :external+sklearn:ref:`scikit-learn 1.8 <release_notes_1_8>` (:issue:`580`).
+- Add support for :external+pandas:ref:`pandas 3.0 <whatsnew_300>` (:issue:`579`).
+- Add support for numpy 2.4 (:issue:`579`).
+
+Bug Fixes
+^^^^^^^^^
+
+- Fix splitting logic in :class:`sksurv.tree.SurvivalTree` and :class:`sksurv.ensemble.RandomSurvivalForest`
+  to correctly handle missing values when using the log-rank criterion. This ensures consistency
+  with scikit-learn 1.8.
+- Fix "cannot resize array" error in Cython extension of :class:`sksurv.svm.MinlipSurvivalAnalysis`
+  by disabling refcheck.
+
+Backwards incompatible changes
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+- Require at least scikit-learn 1.8.0.
+- Require at least pandas 2.2.0.
+- Require at least scipy 1.13.0.
+- Require at least numpy 2.0.0.

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/doc/release_notes.rst

@@ -4,6 +4,7 @@ Release Notes
 .. toctree::
   :maxdepth: 2
 
+  release_notes/v0.27
   release_notes/v0.26
   release_notes/v0.25
   release_notes/v0.24

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/pyproject.toml

@@ -4,12 +4,12 @@ requires = [
     "setuptools-scm>=8",
     "packaging",
     # same as scikit-learn
-    "Cython>=3.0.10",
+    "Cython>=3.1.4",
     # building against numpy 2.x is compatible with numpy 1.x
     "numpy>=2.0.0",
 
     # scikit-learn requirements
-    "scikit-learn~=1.7.0; python_version<='3.14'",
+    "scikit-learn~=1.8.0; python_version<='3.14'",
     "scikit-learn; python_version>'3.14'",
 ]
 build-backend = "setuptools.build_meta"
@@ -46,11 +46,11 @@ dependencies = [
     "ecos",
     "joblib",
     "numexpr",
-    "numpy",
+    "numpy >=2.0.0",
     "osqp >=1.0.2",
-    "pandas >=2.0.0",
-    "scipy >=1.3.2",
-    "scikit-learn >=1.6.1,<1.8",
+    "pandas >=2.2.0",
+    "scipy >=1.13.0",
+    "scikit-learn >=1.8.0,<1.9",
 ]
 dynamic = ["version"]
 
@@ -90,7 +90,7 @@ test = [
 test-nb = [
     "ipykernel",
     "ipython!=8.7.0",
-    "matplotlib~=3.9.0",
+    "matplotlib~=3.10.0",
     "nbformat",
     "nbval>=0.10.0",
     "seaborn~=0.13.2",
@@ -132,25 +132,17 @@ testpaths = ["tests"]
 filterwarnings = [
     # Treat all warnings as errors other than the ignored ones
     "error",
-    # distutils is deprecated in 3.10, scheduled for removal in 3.12
-    "ignore:The distutils package is deprecated:DeprecationWarning",
-    "ignore:Setuptools is replacing distutils",
-    "ignore:distutils Version classes are deprecated.*:DeprecationWarning",
     "ignore:Jupyter is migrating its paths to use standard platformdirs:DeprecationWarning",
-    # added with pandas 2.1
-    "ignore:DataFrame.applymap has been deprecated:FutureWarning",
     # deprecated since Pandas 2.2.0
     "ignore:\\nPyarrow will become a required dependency of pandas in the next major release of pandas.*:DeprecationWarning",
     # deprecated since Python 3.12
     "ignore:datetime\\.datetime\\.utcfromtimestamp\\(\\) is deprecated and scheduled for removal in a future version.*:DeprecationWarning",
     "ignore:ast\\.Num is deprecated and will be removed in Python 3\\.14.*:DeprecationWarning",
     "ignore:Attribute n is deprecated and will be removed in Python 3\\.14.*:DeprecationWarning",
-    # deprecated since NumPy 1.25
-    "ignore:np\\.find_common_type is deprecated.  Please use `np\\.result_type` or `np\\.promote_types`:DeprecationWarning",
-    # deprecated since NumPy 2.0
-    "ignore:`trapz` is deprecated\\. Use `trapezoid` instead.*:DeprecationWarning",
     # deprecated since scipy 1.16.0, scheduled for removal in 1.18
     "ignore:scipy\\.optimize. The `disp` and `iprint` options of the L-BFGS-B solver are deprecated.*:DeprecationWarning",
+    # deprecated since pandas 3.0.0
+    "ignore:The copy keyword is deprecated and will be removed in a future version.*:DeprecationWarning",
 ]
 
 [tool.coverage.run]
@@ -196,7 +188,6 @@ target-version = "py311"
 
 [tool.ruff.lint]
 ignore = ["A005", "C408"]
-ignore-init-module-imports = true
 select = [
     # pycodestyle
     "E",

{scikit_survival-0.26.0 → scikit_survival-0.27.0/scikit_survival.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scikit-survival
-Version: 0.26.0
+Version: 0.27.0
 Summary: Survival analysis built on top of scikit-learn
 Author-email: Sebastian Pölsterl <sebp@k-d-w.org>
 License-Expression: GPL-3.0-or-later
@@ -31,16 +31,16 @@ License-File: COPYING
 Requires-Dist: ecos
 Requires-Dist: joblib
 Requires-Dist: numexpr
-Requires-Dist: numpy
+Requires-Dist: numpy>=2.0.0
 Requires-Dist: osqp>=1.0.2
-Requires-Dist: pandas>=2.0.0
-Requires-Dist: scipy>=1.3.2
-Requires-Dist: scikit-learn<1.8,>=1.6.1
+Requires-Dist: pandas>=2.2.0
+Requires-Dist: scipy>=1.13.0
+Requires-Dist: scikit-learn<1.9,>=1.8.0
 Dynamic: license-file
 
 |License| |Docs| |DOI|
 
-|build-tests| |build-windows| |Codecov| |Codacy|
+|build-tests| |Codecov| |Codacy|
 
 ***************
 scikit-survival
@@ -76,10 +76,10 @@ Requirements
 - ecos
 - joblib
 - numexpr
-- numpy
+- numpy 2.0.0 or later
 - osqp
-- pandas 2.0.0 or later
-- scikit-learn 1.6 or 1.7
+- pandas 2.2.0 or later
+- scikit-learn 1.8
 - scipy
 - C/C++ compiler
 
@@ -178,8 +178,4 @@ Please cite the following paper if you are using **scikit-survival**.
   :target: https://github.com/sebp/scikit-survival/actions?query=workflow%3Atests+branch%3Amaster
   :alt: GitHub Actions Tests Status
 
-.. |build-windows| image:: https://ci.appveyor.com/api/projects/status/github/sebp/scikit-survival?branch=master&svg=true
-   :target: https://ci.appveyor.com/project/sebp/scikit-survival
-   :alt: Windows Build Status on AppVeyor
-
 .. _survival analysis: https://en.wikipedia.org/wiki/Survival_analysis

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/scikit_survival.egg-info/SOURCES.txt

@@ -53,6 +53,7 @@ doc/release_notes/v0.23.rst
 doc/release_notes/v0.24.rst
 doc/release_notes/v0.25.rst
 doc/release_notes/v0.26.rst
+doc/release_notes/v0.27.rst
 doc/release_notes/v0.3.rst
 doc/release_notes/v0.4.rst
 doc/release_notes/v0.5.rst

scikit_survival-0.27.0/scikit_survival.egg-info/requires.txt

@@ -0,0 +1,8 @@
+ecos
+joblib
+numexpr
+numpy>=2.0.0
+osqp>=1.0.2
+pandas>=2.2.0
+scipy>=1.13.0
+scikit-learn<1.9,>=1.8.0

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/setup.py

@@ -18,7 +18,7 @@ import sys
 from packaging.version import Version
 from setuptools import Command, Extension, setup
 
-CYTHON_MIN_VERSION = Version("3.0.10")
+CYTHON_MIN_VERSION = Version("3.1.4")
 
 
 # adapted from bottleneck's setup.py

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/sksurv/io/arffwrite.py

@@ -172,7 +172,7 @@ def _write_data(data, fp):
             return "?"
         return str(x)
 
-    data = data.applymap(to_str)
+    data = data.map(to_str)
     n_rows = data.shape[0]
     for i in range(n_rows):
         str_values = list(data.iloc[i, :].apply(_check_str_array))

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/sksurv/metrics.py

@@ -510,7 +510,7 @@ def cumulative_dynamic_auc(survival_train, survival_test, estimate, times, tied_
             # to make sure that the curve starts at (0, 0)
             tp_no_ties = np.r_[0, tp_no_ties]
             fp_no_ties = np.r_[0, fp_no_ties]
-            scores[i] = np.trapz(tp_no_ties, fp_no_ties)
+            scores[i] = np.trapezoid(tp_no_ties, fp_no_ties)
 
     if n_times == 1:
         mean_auc = scores[0]
@@ -780,7 +780,7 @@ def integrated_brier_score(survival_train, survival_test, estimate, times):
         raise ValueError("At least two time points must be given")
 
     # Computing the IBS
-    ibs_value = np.trapz(brier_scores, times) / (times[-1] - times[0])
+    ibs_value = np.trapezoid(brier_scores, times) / (times[-1] - times[0])
 
     return ibs_value
 

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/sksurv/svm/_minlip.pyx

@@ -64,11 +64,13 @@ def create_difference_matrix(const cnp.npy_uint8[:] event,
     new_dim.ptr = &k
     new_dim.len = 1
 
-    ret = cnp.PyArray_Resize(values, &new_dim, 1, cnp.NPY_CORDER)
+    # disable refcheck, we are certain the array is not referenced by other arrays
+    cdef int refcheck = 0
+    ret = cnp.PyArray_Resize(values, &new_dim, refcheck, cnp.NPY_CORDER)
     if ret is not None:  # returns NULL on error
         return
 
-    ret = cnp.PyArray_Resize(columns, &new_dim, 1, cnp.NPY_CORDER)
+    ret = cnp.PyArray_Resize(columns, &new_dim, refcheck, cnp.NPY_CORDER)
     if ret is not None:  # returns NULL on error
         return
 

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/sksurv/testing.py

@@ -119,7 +119,7 @@ class FixtureParameterFactory:
         return cases
 
 
-def check_module_minimum_version(module, version_str):
+def check_module_minimum_version(module, min_version_str, max_version_str=None):
     """
     Check whether a module of a specified minimum version is available.
 
@@ -127,8 +127,10 @@ def check_module_minimum_version(module, version_str):
     ----------
     module : str
         Name of the module.
-    version_str : str
+    min_version_str : str
         Minimum version of the module.
+    max_version_str : str, optional
+        Maximum version of the module (excluding).
 
     Returns
     -------
@@ -137,14 +139,17 @@ def check_module_minimum_version(module, version_str):
     """
     try:
         module_version = parse(version(module))
-        required_version = parse(version_str)
-        return module_version >= required_version
+        required_min_version = parse(min_version_str)
+        if max_version_str is None:
+            return module_version >= required_min_version
+        required_max_version = parse(max_version_str)
+        return required_min_version <= module_version < required_max_version
     except PackageNotFoundError:  # pragma: no cover
         return False
 
 
 def get_pandas_infer_string_context():
-    if check_module_minimum_version("pandas", "2.3.0"):
+    if check_module_minimum_version("pandas", "2.3.0", "3.0.0"):
         return (
             pytest.param(pd.option_context("future.infer_string", False), id="infer_string=False"),
             pytest.param(pd.option_context("future.infer_string", True), id="infer_string=True"),

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/sksurv/tree/_criterion.pyx

@@ -144,8 +144,12 @@ cdef class LogrankCriterion(Criterion):
         intp_t n_unique_times
         intp_t nbytes
         RisksetCounter riskset_total
+        RisksetCounter riskset_missing
         float64_t * weighted_n_events_left
         float64_t * weighted_delta_n_at_risk_left
+        float64_t * weighted_n_events_missing
+        float64_t * weighted_delta_n_at_risk_missing
+        float64_t weighted_n_missing
         intp_t * samples_time_idx
 
     def __cinit__(self, intp_t n_outputs, intp_t n_samples, const float64_t[::1] unique_times, const cnp.npy_bool[::1] is_event_time):
@@ -164,16 +168,22 @@ cdef class LogrankCriterion(Criterion):
         self.weighted_n_node_samples = 0.0
         self.weighted_n_left = 0.0
         self.weighted_n_right = 0.0
+        self.weighted_n_missing = 0.0
 
         self.riskset_total = RisksetCounter(unique_times)
+        self.riskset_missing = RisksetCounter(unique_times)
         self.weighted_delta_n_at_risk_left = <float64_t *> malloc(self.nbytes)
         self.weighted_n_events_left = <float64_t *> malloc(self.nbytes)
+        self.weighted_delta_n_at_risk_missing = <float64_t *> malloc(self.nbytes)
+        self.weighted_n_events_missing = <float64_t *> malloc(self.nbytes)
         self.samples_time_idx = <intp_t *> malloc(n_samples * sizeof(intp_t))
 
     def __dealloc__(self):
         """Destructor."""
         free(self.weighted_delta_n_at_risk_left)
         free(self.weighted_n_events_left)
+        free(self.weighted_delta_n_at_risk_missing)
+        free(self.weighted_n_events_missing)
         free(self.samples_time_idx)
 
     def __reduce__(self):
@@ -238,6 +248,38 @@ cdef class LogrankCriterion(Criterion):
         self.pos = self.end
         return 0
 
+    cdef void init_missing(self, intp_t n_missing) noexcept nogil:
+        """Initialize the criterion for the missing samples."""
+        cdef:
+            intp_t i
+            intp_t idx
+            float64_t event
+            intp_t time_idx
+            float64_t w = 1.0
+            const float64_t[:] sample_weight = self.sample_weight
+            const intp_t[:] samples = self.sample_indices
+            const float64_t[:, ::1] y = self.y
+
+        self.n_missing = n_missing
+        self.weighted_n_missing = 0.0
+        memset(self.weighted_delta_n_at_risk_missing, 0, self.nbytes)
+        memset(self.weighted_n_events_missing, 0, self.nbytes)
+
+        if n_missing > 0:
+            for i in range(self.end - n_missing, self.end):
+                idx = samples[i]
+                event = y[idx, 1]
+                time_idx = self.samples_time_idx[idx]
+
+                if sample_weight is not None:
+                    w = sample_weight[idx]
+
+                self.weighted_delta_n_at_risk_missing[time_idx] += w
+                if event != 0.0:
+                    self.weighted_n_events_missing[time_idx] += w
+
+                self.weighted_n_missing += w
+
     cdef int update(self, intp_t new_pos) except -1 nogil:
         """Updated statistics by moving samples[pos:new_pos] to the left."""
         cdef:
@@ -292,15 +334,25 @@ cdef class LogrankCriterion(Criterion):
             intp_t i
             float64_t weighted_at_risk = self.weighted_n_left
             float64_t events
+            float64_t delta_at_risk
             float64_t total_at_risk
             float64_t total_events
             float64_t ratio
             float64_t v
             float64_t denom = 0.0
             float64_t numer = 0.0
+            int missing_go_to_left = self.missing_go_to_left
+
+        if missing_go_to_left:
+            weighted_at_risk += self.weighted_n_missing
 
         for i in range(self.n_unique_times):
             events = self.weighted_n_events_left[i]
+            delta_at_risk = self.weighted_delta_n_at_risk_left[i]
+            if missing_go_to_left:
+                events += self.weighted_n_events_missing[i]
+                delta_at_risk += self.weighted_delta_n_at_risk_missing[i]
+
             self.riskset_total.at(i, &total_at_risk, &total_events)
 
             if total_at_risk == 0:
@@ -312,7 +364,7 @@ cdef class LogrankCriterion(Criterion):
                 denom += ratio * (1.0 - ratio) * v
 
             # Update number of samples at risk for next bigger timepoint
-            weighted_at_risk -= self.weighted_delta_n_at_risk_left[i]
+            weighted_at_risk -= delta_at_risk
 
         if denom != 0.0:
             # absolute value is the measure of node separation

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/tests/conftest.py

@@ -16,7 +16,7 @@ DataSet = namedtuple("DataSet", ["x", "y"])
 DataSetWithNames = namedtuple("DataSetWithNames", ["x", "y", "names", "x_data_frame"])
 SparseDataSet = namedtuple("SparseDataSet", ["x_dense", "x_sparse", "y"])
 
-if Version(pd.__version__) >= Version("2.3.0"):
+if Version("2.3.0") <= Version(pd.__version__) < Version("3.0.0"):
     pd.set_option("mode.copy_on_write", True)
 
 

{scikit_survival-0.26.0 → scikit_survival-0.27.0}/tests/test_forest.py

@@ -41,20 +41,20 @@ def test_fit_predict(make_whas500, forest_cls, expected_c):
 
 
 @pytest.mark.parametrize(
-    "forest_cls,expected_cindex", [(ExtraSurvivalTrees, 0.7379332685455134), (RandomSurvivalForest, 0.732102364755426)]
+    "forest_cls,expected_cindex", [(ExtraSurvivalTrees, 0.7644962747003563), (RandomSurvivalForest, 0.7638483965014577)]
 )
 def test_fit_missing_values(make_whas500, forest_cls, expected_cindex):
     whas500 = make_whas500(to_numeric=True)
 
-    rng = np.random.default_rng(42)
-    mask = rng.binomial(n=1, p=0.15, size=whas500.x.shape)
-    mask = mask.astype(bool)
     X = whas500.x.copy()
-    X[mask] = np.nan
-
     X_train, y_train = X[:400], whas500.y[:400]
     X_test, y_test = X[400:], whas500.y[400:]
 
+    rng = np.random.default_rng(42)
+    mask = rng.binomial(n=1, p=0.15, size=X_train.shape)
+    mask = mask.astype(bool)
+    X_train[mask] = np.nan
+
     forest = forest_cls(random_state=42)
     forest.fit(X_train, y_train)