Update recurrent_utils.py

Improved runtime efficiency of handle_xicn function.

surpyval/utils/recurrent_utils.py

@@ -54,14 +54,6 @@ def handle_xicn(x, i=None, c=None, n=None, Z=None, as_recurrent_data=True):
             "Counts greater than 1 must be intervally or left censored"
         )
 
-    # Check that if censored, it is the highest value
-    for ii in set(i):
-        ci = c[i == ii]
-        if len(ci[ci == 1]) > 1:
-            raise ValueError(f"Item {ii} is right censored more than once.")
-        if len(ci[ci == -1]) > 1:
-            raise ValueError(f"Item {ii} is left censored more than once.")
-
     # sort by item and x
     if x.ndim == 2:
         # Order 2D by the midpoint
@@ -74,17 +66,36 @@ def handle_xicn(x, i=None, c=None, n=None, Z=None, as_recurrent_data=True):
     if Z is not None:
         Z = Z[sort_order]
 
-    # Check that the x values for each item are monotonically increasing
-    for ii in set(i):
-        xi = x[i == ii]
-        ci = c[i == ii]
-        if xi.ndim == 2:
-            for first, second in zip(xi[:-1], xi[1:]):
-                if first[1] > second[0]:
-                    raise ValueError(f"Item {ii} has overlapping intervals")
-        else:
-            if np.any(np.diff(xi) < 0):
-                raise ValueError(f"Item {ii} has non-monotonic x values")
+    unique_i, idx = np.unique(i, return_index=True)
+    censoring_by_i = np.split(c, idx)[1:]
+
+    for ii, arr in zip(unique_i, censoring_by_i):
+        if 1 in arr:
+            if (arr == 1).sum() > 1:
+                raise ValueError(
+                    f"Item {ii} has more than one right censored time"
+                )
+            if arr[-1] != 1:
+                raise ValueError(
+                    f"Item {ii} has right censored event which is not the last"
+                )
+        if -1 in arr:
+            if (arr == -1).sum() > 1:
+                raise ValueError(
+                    f"Item {ii} has more than one left censored event"
+                )
+            if arr[0] != -1:
+                raise ValueError(
+                    f"Item {ii} has left censored event that is not the first"
+                )
+
+    if x.ndim == 2:
+        times_by_i = np.split(x, idx)[1:]
+        for ii, arr in zip(unique_i, times_by_i):
+            starts = arr[1:][:, 0]
+            ends = arr[:-1][:, 1]
+            if (ends > starts).any():
+                raise ValueError(f"Item {ii} has overlapping intervals")
 
     if as_recurrent_data:
         data = RecurrentEventData(x, i, c, n)