nlogn concordance index algorithm (first pass)

lifelines/estimation.py

@@ -1307,12 +1307,10 @@ def summary(self):
         df['upper %.2f' % self.alpha] = self.confidence_intervals_.ix['upper-bound'].values
         return df
 
-    def print_summary(self, c_index=True):
+    def print_summary(self):
         """
         Print summary statistics describing the fit.
 
-        c_index: If set to False, will not print the concordance index
-
         """
         df = self.summary
         # Significance codes last
@@ -1327,11 +1325,10 @@ def print_summary(self, c_index=True):
         print('---')
         print("Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 ",
               end='\n\n')
-        if c_index:
-            print("Concordance = {:.3f}"
-                  .format(concordance_index(self.durations,
-                                            -self.predict_partial_hazard(self.data).values.ravel(),
-                                            self.event_observed)))
+        print("Concordance = {:.3f}"
+              .format(concordance_index(self.durations,
+                                        -self.predict_partial_hazard(self.data).values.ravel(),
+                                        self.event_observed)))
         return
 
     def predict_partial_hazard(self, X):

lifelines/tests/test_estimation.py

@@ -618,7 +618,7 @@ def test_print_summary(self):
 ---
 Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
 
-Concordance = 0.644""".strip().split()
+Concordance = 0.640""".strip().split()
             for i in [0, 1, 2, -2, -1]:
                 assert output[i] == expected[i]
         finally:

lifelines/tests/test_utils.py

@@ -10,13 +10,10 @@
 from ..estimation import CoxPHFitter
 from ..datasets import (load_regression_dataset, load_larynx,
                         load_waltons, load_rossi)
-from .._utils.cindex import concordance_index as slow_cindex
 from lifelines import utils
-try:
-    from .._utils._cindex import concordance_index as fast_cindex
-except ImportError:
-    # If code has not been compiled.
-    fast_cindex = None
+from lifelines.utils import _concordance_index as fast_cindex
+from lifelines.utils import _naive_concordance_index as slow_cindex
+from lifelines.utils import _BTree as BTree
 
 
 def test_ridge_regression_with_penalty_is_less_than_without_penalty():
@@ -350,8 +347,38 @@ def test_survival_table_from_events_raises_value_error_if_too_early_births():
         utils.survival_table_from_events(T, C, min_obs)
 
 
-@pytest.mark.skipif(fast_cindex is None, reason='extensions not compiled')
-def test_concordance_index_py_is_same_as_native():
+def test_btree():
+    t = BTree(np.arange(10))
+    for i in range(10):
+        assert t.rank(i) == (0, 0)
+
+    assert len(t) == 0
+    t.insert(5)
+    t.insert(6)
+    t.insert(6)
+    t.insert(0)
+    t.insert(9)
+    assert len(t) == 5
+
+    assert t.rank(0) == (0, 1)
+    assert t.rank(0.5) == (1, 0)
+    assert t.rank(4.5) == (1, 0)
+    assert t.rank(5) == (1, 1)
+    assert t.rank(5.5) == (2, 0)
+    assert t.rank(6) == (2, 2)
+    assert t.rank(6.5) == (4, 0)
+    assert t.rank(8.5) == (4, 0)
+    assert t.rank(9) == (4, 1)
+    assert t.rank(9.5) == (5, 0)
+
+    for i in range(1, 32):
+        BTree(np.arange(i))
+
+    with pytest.raises(ValueError):
+        # This has to go last since it screws up the counts
+        t.insert(5.5)
+
+def test_concordance_index_fast_is_same_as_slow():
     size = 100
     T = np.random.normal(size=size)
     P = np.random.normal(size=size)