FIX weights computation with ties in IsotonicRegression (scikit-learn…

…#9484)

doc/whats_new/v0.20.rst

@@ -16,6 +16,7 @@ occurs due to changes in the modelling logic (bug fixes or enhancements), or in
 random sampling procedures.
 
 - :class:`decomposition.IncrementalPCA` in Python 2 (bug fix)
+- :class:`isotonic.IsotonicRegression` (bug fix)
 
 Details are listed in the changelog below.
 
@@ -68,6 +69,13 @@ Linear, kernelized and related models
 Bug fixes
 .........
 
+Classifiers and regressors
+
+- Fixed a bug in :class:`isotonic.IsotonicRegression` which incorrectly
+  combined weights when fitting a model to data involving points with
+  identical X values.
+  :issue:`9432` by :user:`Dallas Card <dallascard>`
+
 Decomposition, manifold learning and clustering
 
 - Fix for uninformative error in :class:`decomposition.IncrementalPCA`:

sklearn/_isotonic.pyx

@@ -100,7 +100,7 @@ def _make_unique(np.ndarray[dtype=np.float64_t] X,
         if x != current_x:
             # next unique value
             x_out[i] = current_x
-            weights_out[i] = current_weight / current_count
+            weights_out[i] = current_weight
             y_out[i] = current_y / current_weight
             i += 1
             current_x = x
@@ -113,6 +113,6 @@ def _make_unique(np.ndarray[dtype=np.float64_t] X,
             current_count += 1
 
     x_out[i] = current_x
-    weights_out[i] = current_weight / current_count
+    weights_out[i] = current_weight
     y_out[i] = current_y / current_weight
     return x_out, y_out, weights_out

sklearn/tests/test_isotonic.py

@@ -166,6 +166,30 @@ def test_isotonic_regression_ties_secondary_():
     assert_array_almost_equal(ir.fit_transform(x, y), y_true, 4)
 
 
+def test_isotonic_regression_with_ties_in_differently_sized_groups():
+    """
+    Non-regression test to handle issue 9432:
+    https://github.com/scikit-learn/scikit-learn/issues/9432
+
+    Compare against output in R:
+    > library("isotone")
+    > x <- c(0, 1, 1, 2, 3, 4)
+    > y <- c(0, 0, 1, 0, 0, 1)
+    > res1 <- gpava(x, y, ties="secondary")
+    > res1$x
+
+    `isotone` version: 1.1-0, 2015-07-24
+    R version: R version 3.3.2 (2016-10-31)
+    """
+    x = np.array([0, 1, 1, 2, 3, 4])
+    y = np.array([0, 0, 1, 0, 0, 1])
+    y_true = np.array([0., 0.25, 0.25, 0.25, 0.25, 1.])
+    ir = IsotonicRegression()
+    ir.fit(x, y)
+    assert_array_almost_equal(ir.transform(x), y_true)
+    assert_array_almost_equal(ir.fit_transform(x, y), y_true)
+
+
 def test_isotonic_regression_reversed():
     y = np.array([10, 9, 10, 7, 6, 6.1, 5])
     y_ = IsotonicRegression(increasing=False).fit_transform(