v0.18.3 (#627)

* bump version

* make this test more robust

Author: CamDavidsonPilon

CHANGELOG.md

@@ -1,6 +1,6 @@
 ### Changelogs
 
-### 0.18.3 (dev)
+### 0.18.3 
  - Some performance improvements to parametric univariate models.
  - Suppressing some irrelevant NumPy and autograd warnings, so lifeline warnings are more noticeable. 
  - Improved some warning and error messages. 

docs/conf.py

@@ -60,7 +60,7 @@
 #
 # The short X.Y version.
 
-version = "0.18.2"
+version = "0.18.3"
 # The full version, including dev info
 release = version
 

lifelines/fitters/__init__.py

@@ -4,7 +4,8 @@
 from functools import wraps
 import sys
 import warnings
-#pylint: disable=wrong-import-position
+
+# pylint: disable=wrong-import-position
 
 warnings.simplefilter(action="ignore", category=FutureWarning)
 from textwrap import dedent
@@ -238,13 +239,13 @@ def __init__(self, *args, **kwargs):
         if "alpha" in self._fitted_parameter_names:
             raise NameError("'alpha' in _fitted_parameter_names is a lifelines reserved word. Try 'alpha_' instead.")
 
-    def _check_cumulative_hazard_is_monotone_and_positive(self):
-        test_times = np.linspace(0.01, 100, 15)
+    def _check_cumulative_hazard_is_monotone_and_positive(self, durations):
         class_name = self.__class__.__name__
 
-        cumulative_hazard = self._cumulative_hazard(self._initial_values, test_times)
+        cumulative_hazard = self._cumulative_hazard(self._initial_values, durations)
         if not np.all(cumulative_hazard > 0):
-            raise StatisticalWarning(dedent(
+            warnings.warn(
+                dedent(
                     """\
                 Cumulative hazard is not strictly positive. For example, try:
                 
@@ -253,11 +254,17 @@ def _check_cumulative_hazard_is_monotone_and_positive(self):
                 >>> fitter._cumulative_hazard(fitter._initial_values, test_times)
 
                 This may harm convergence, or return nonsensical results.
-            """.format(class_name)))
+            """.format(
+                        class_name
+                    )
+                ),
+                StatisticalWarning,
+            )
 
-        derivative_of_cumulative_hazard = self._hazard(self._initial_values, test_times)
+        derivative_of_cumulative_hazard = self._hazard(self._initial_values, durations)
         if not np.all(derivative_of_cumulative_hazard >= 0):
-            raise StatisticalWarning(dedent(
+            warnings.warn(
+                dedent(
                     """\
                 Cumulative hazard is not strictly non-decreasing. For example, try:
                 
@@ -266,7 +273,12 @@ def _check_cumulative_hazard_is_monotone_and_positive(self):
                 >>> fitter._hazard(fitter._initial_values, test_times)
 
                 This may harm convergence, or return nonsensical results.
-            """.format(class_name)))
+            """.format(
+                        class_name
+                    )
+                ),
+                StatisticalWarning,
+            )
 
     def _initial_values_from_bounds(self):
         for (lb, ub) in self._bounds:
@@ -472,16 +484,15 @@ def fit(
         if event_observed is not None:
             check_nans_or_infs(event_observed)
 
-        self._check_cumulative_hazard_is_monotone_and_positive()
-
-
         self.durations = np.asarray(durations, dtype=float)
         # check for negative or 0 durations - these are not allowed in a weibull model.
         if np.any(self.durations <= 0):
             raise ValueError(
                 "This model does not allow for non-positive durations. Suggestion: add a small positive value to zero elements."
             )
 
+        self._check_cumulative_hazard_is_monotone_and_positive(self.durations)
+
         self.event_observed = (
             np.asarray(event_observed, dtype=int) if event_observed is not None else np.ones_like(self.durations)
         )

lifelines/version.py

@@ -1,4 +1,4 @@
 # -*- coding: utf-8 -*-
 from __future__ import unicode_literals
 
-__version__ = "0.18.2"
+__version__ = "0.18.3"

tests/test_estimation.py

@@ -32,7 +32,7 @@
     StatisticalWarning,
 )
 
-from lifelines.fitters import BaseFitter
+from lifelines.fitters import BaseFitter, ParametericUnivariateFitter
 
 from lifelines import (
     WeibullFitter,
@@ -180,6 +180,27 @@ def test_parametric_univarite_fitters_can_print_summary(
             f.summary
             f.print_summary()
 
+    def test_warnings_for_problematic_cumulative_hazards(self):
+        class NegativeFitter(ParametericUnivariateFitter):
+
+            _fitted_parameter_names = ["a"]
+
+            def _cumulative_hazard(self, params, times):
+                return params[0] * (times - 0.4)
+
+        class DecreasingFitter(ParametericUnivariateFitter):
+
+            _fitted_parameter_names = ["a"]
+
+            def _cumulative_hazard(self, params, times):
+                return params[0] * 1 / times
+
+        with pytest.warns(StatisticalWarning, match="positive") as w:
+            NegativeFitter().fit([0.01, 0.5, 10.0, 20.0])
+
+        with pytest.warns(StatisticalWarning, match="non-decreasing") as w:
+            DecreasingFitter().fit([0.01, 0.5, 10.0, 20])
+
 
 class TestUnivariateFitters:
     def test_univarite_fitters_with_survival_function_have_conditional_time_to_(
@@ -451,7 +472,7 @@ def test_lnf_inference(self, lnf):
 
         assert abs(mu - lnf.mu_) < 0.05
         assert abs(sigma - lnf.sigma_) < 0.05
-        assert abs(lnf.median_ - np.percentile(X, 50)) < 0.05
+        assert abs(lnf.median_/np.percentile(X, 50) - 1) < 0.05
 
     def test_lnf_inference_with_large_sigma(self, lnf):
         N = 250000