Merge 9cf84d0 into fab70f7

lifelines/fitters/__init__.py

@@ -620,10 +620,10 @@ def print_summary(self, decimals=2, **kwargs):
             multiple outputs.
 
         """
-        justify = utils.string_justify(18)
+        justify = utils.string_justify(25)
         print(self)
-        print("{} = {}".format(justify("number of observations"), self.weights.sum()))
-        print("{} = {}".format(justify("number of events observed"), self.weights[self.event_observed > 0].sum()))
+        print("{} = {:g}".format(justify("number of observations"), self.weights.sum()))
+        print("{} = {:g}".format(justify("number of events observed"), self.weights[self.event_observed > 0].sum()))
         print("{} = {:.{prec}f}".format(justify("log-likelihood"), self.log_likelihood_, prec=decimals))
         print(
             "{} = {}".format(
@@ -1721,7 +1721,7 @@ def print_summary(self, decimals=2, **kwargs):
         """
 
         # Print information about data first
-        justify = utils.string_justify(18)
+        justify = utils.string_justify(25)
         print(self)
         if self.event_col:
             print("{} = '{}'".format(justify("event col"), self.event_col))
@@ -1733,8 +1733,8 @@ def print_summary(self, decimals=2, **kwargs):
         if self.robust:
             print("{} = {}".format(justify("robust variance"), True))
 
-        print("{} = {}".format(justify("number of observations"), self.weights.sum()))
-        print("{} = {}".format(justify("number of events observed"), self.weights[self.event_observed > 0].sum()))
+        print("{} = {:g}".format(justify("number of observations"), self.weights.sum()))
+        print("{} = {:g}".format(justify("number of events observed"), self.weights[self.event_observed > 0].sum()))
         print("{} = {:.{prec}f}".format(justify("log-likelihood"), self.log_likelihood_, prec=decimals))
         print("{} = {}".format(justify("time fit was run"), self._time_fit_was_called))
 
@@ -1879,7 +1879,7 @@ def predict_cumulative_hazard(self, df, times=None, conditional_after=None):
 
         """
         df = df.copy().astype(float)
-        times = utils.coalesce(times, self.timeline, np.unique(self.durations))
+        times = utils.coalesce(times, self.timeline)
         times = np.atleast_1d(times).astype(float)
 
         if isinstance(df, pd.Series):
@@ -2946,7 +2946,7 @@ def predict_cumulative_hazard(self, df, ancillary_df=None, times=None, condition
         predict_percentile, predict_expectation, predict_survival_function
         """
         df = df.copy().astype(float)
-        times = utils.coalesce(times, self.timeline, np.unique(self.durations))
+        times = utils.coalesce(times, self.timeline)
         times = np.atleast_1d(times).astype(float)
 
         if isinstance(df, pd.Series):

lifelines/fitters/coxph_fitter.py

@@ -1267,7 +1267,7 @@ def print_summary(self, decimals=2, **kwargs):
         """
 
         # Print information about data first
-        justify = string_justify(18)
+        justify = string_justify(25)
         print(self)
         print("{} = '{}'".format(justify("duration col"), self.duration_col))
 
@@ -1288,8 +1288,8 @@ def print_summary(self, decimals=2, **kwargs):
         if self.penalizer > 0:
             print("{} = {}".format(justify("penalizer"), self.penalizer))
 
-        print("{} = {}".format(justify("number of observations"), self.weights.sum()))
-        print("{} = {}".format(justify("number of events observed"), self.weights[self.event_observed > 0].sum()))
+        print("{} = {:g}".format(justify("number of observations"), self.weights.sum()))
+        print("{} = {:g}".format(justify("number of events observed"), self.weights[self.event_observed > 0].sum()))
         print("{} = {:.{prec}f}".format(justify("partial log-likelihood"), self.log_likelihood_, prec=decimals))
         print("{} = {}".format(justify("time fit was run"), self._time_fit_was_called))
 

tests/test_estimation.py

@@ -295,6 +295,14 @@ def test_parameteric_models_all_can_do_interval_censoring(self, known_parametric
         for fitter in known_parametric_univariate_fitters:
             f = fitter().fit_interval_censoring(df["left"], df["right"])
 
+    def test_parameteric_models_all_can_do_interval_censoring_with_prediction(
+        self, known_parametric_univariate_fitters
+    ):
+        df = load_diabetes()
+        for fitter in known_parametric_univariate_fitters:
+            f = fitter().fit_interval_censoring(df["left"], df["right"])
+            f.predict(3.0)
+
     def test_parameteric_models_fail_if_passing_in_bad_event_data(self, known_parametric_univariate_fitters):
         df = load_diabetes()
         for fitter in known_parametric_univariate_fitters:
@@ -2195,7 +2203,6 @@ def test_weibull_interval_censoring_inference_on_known_R_output(self, aft):
         df["E"] = df["left"] == df["right"]
 
         aft.fit_interval_censoring(df, "left", "right", "E")
-        print(aft.summary)
         npt.assert_allclose(aft.summary.loc[("lambda_", "gender"), "coef"], 0.04576, rtol=1e-3)
         npt.assert_allclose(aft.summary.loc[("lambda_", "_intercept"), "coef"], np.log(18.31971), rtol=1e-4)
         npt.assert_allclose(aft.summary.loc[("rho_", "_intercept"), "coef"], np.log(2.82628), rtol=1e-4)
@@ -2228,7 +2235,6 @@ def test_aft_weibull_with_weights(self, rossi, aft):
         npt.assert_allclose(aft.summary.loc[("lambda_", "_intercept"), "coef"], 4.93041526, rtol=1e-2)
         npt.assert_allclose(aft.summary.loc[("rho_", "_intercept"), "coef"], 0.28612353, rtol=1e-4)
 
-    @pytest.mark.xfail()
     def test_aft_weibull_with_ancillary_model_and_with_weights(self, rossi):
         """
         library('flexsurv')
@@ -2237,10 +2243,19 @@ def test_aft_weibull_with_ancillary_model_and_with_weights(self, rossi):
         """
         wf = WeibullAFTFitter(penalizer=0).fit(rossi, "week", "arrest", weights_col="age", ancillary_df=rossi[["prio"]])
 
-        npt.assert_allclose(wf.summary.loc[("lambda_", "fin"), "coef"], 0.360792647, rtol=1e-3)
-        npt.assert_allclose(wf.summary.loc[("rho_", "prio"), "coef"], -0.025505680, rtol=1e-4)
-        npt.assert_allclose(wf.summary.loc[("lambda_", "_intercept"), "coef"], 4.707300215, rtol=1e-2)
-        npt.assert_allclose(wf.summary.loc[("rho_", "_intercept"), "coef"], 0.367701670, rtol=1e-4)
+        npt.assert_allclose(wf.summary.loc[("lambda_", "fin"), "coef"], 0.39347, rtol=1e-3)
+        npt.assert_allclose(wf.summary.loc[("lambda_", "_intercept"), "coef"], np.log(140.55112), rtol=1e-2)
+        npt.assert_allclose(wf.summary.loc[("rho_", "_intercept"), "coef"], np.log(1.25981), rtol=1e-4)
+        npt.assert_allclose(wf.summary.loc[("rho_", "prio"), "coef"], 0.01485, rtol=1e-4)
+
+    def test_aft_weibull_can_do_interval_prediction(self, aft):
+        # https://github.com/CamDavidsonPilon/lifelines/issues/839
+        df = load_diabetes()
+        df["gender"] = df["gender"] == "male"
+        df["E"] = df["left"] == df["right"]
+
+        aft.fit_interval_censoring(df, "left", "right", "E")
+        aft.predict_survival_function(df)
 
 
 class TestCoxPHFitter: