Add pad to stat_ecdf

closes #725

doc/changelog.qmd

@@ -10,6 +10,10 @@ title: Changelog
 - Requires python >= 3.9
 - The name of the calculated aesthetic of
   [](:class:`~plotnine.stats.stat_function`) changed from `y` to `fx`.
+- [](:class:`~plotnine.stats.stat_ecdf`) has gained the `pad` parameter. The
+  default is set to `True`, which pads the domain with `-inf` and `inf` so
+  that the ECDF does not have discontinuities at the extremes. To get the
+  behaviour, set `pad` to `False`. ({{< issue 725 >}})
 
 ### New
 

plotnine/stats/stat_ecdf.py

@@ -19,6 +19,9 @@ class stat_ecdf(stat):
     n  : int, default=None
         This is the number of points to interpolate with.
         If :py:`None`, do not interpolate.
+    pad : bool, default=True
+        If True, pad the domain with `-inf` and `+inf` so that
+        ECDF does not have discontinuities at the extremes.
 
     See Also
     --------
@@ -42,6 +45,7 @@ class stat_ecdf(stat):
         "position": "identity",
         "na_rm": False,
         "n": None,
+        "pad": True,
     }
     DEFAULT_AES = {"y": after_stat("ecdf")}
     CREATES = {"ecdf"}
@@ -56,6 +60,9 @@ def compute_group(cls, data, scales, **params):
         else:
             x = np.linspace(data["x"].min(), data["x"].max(), params["n"])
 
+        if params["pad"]:
+            x = np.hstack([-np.inf, x, np.inf])
+
         ecdf = ECDF(data["x"])(x)
         res = pd.DataFrame({"x": x, "ecdf": ecdf})
         return res

tests/test_stat_ecdf.py

@@ -12,6 +12,12 @@ def test_ecdf():
     assert p == "ecdf"
 
 
+def test_ecdf_no_pad():
+    p = ggplot(data, aes("x")) + stat_ecdf(size=2, pad=False)
+
+    assert p == "ecdf_no_pad"
+
+
 def test_computed_y_column():
     p = (
         ggplot(data, aes("x"))