matplotlib
diff --git a/‎doc/api/next_api_changes/behavior/27589-DS.rst
Copy file name to clipboard
+5Lines changed: 5 additions & 0 deletions b/‎doc/api/next_api_changes/behavior/27589-DS.rst
Copy file name to clipboard
+5Lines changed: 5 additions & 0 deletions
diff --git a/‎lib/matplotlib/colors.py
Copy file name to clipboardExpand all lines: lib/matplotlib/colors.py
+20-12Lines changed: 20 additions & 12 deletions b/‎lib/matplotlib/colors.py
Copy file name to clipboardExpand all lines: lib/matplotlib/colors.py
+20-12Lines changed: 20 additions & 12 deletions
diff --git a/‎lib/matplotlib/tests/test_colorbar.py
Copy file name to clipboardExpand all lines: lib/matplotlib/tests/test_colorbar.py
+4-4Lines changed: 4 additions & 4 deletions b/‎lib/matplotlib/tests/test_colorbar.py
Copy file name to clipboardExpand all lines: lib/matplotlib/tests/test_colorbar.py
+4-4Lines changed: 4 additions & 4 deletions
diff --git a/‎lib/matplotlib/tests/test_colors.py
Copy file name to clipboardExpand all lines: lib/matplotlib/tests/test_colors.py
+15-2Lines changed: 15 additions & 2 deletions b/‎lib/matplotlib/tests/test_colors.py
Copy file name to clipboardExpand all lines: lib/matplotlib/tests/test_colors.py
+15-2Lines changed: 15 additions & 2 deletions
@@ -0,0 +1,5 @@
+PowerNorm no longer clips values below vmin
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+When ``clip=False`` is set (the default) on `~matplotlib.colors.PowerNorm`,
+values below ``vmin`` are now linearly normalised. Previously they were clipped
+to zero. This fixes issues with the display of colorbars associated with a power norm.
@@ -1953,10 +1953,10 @@ class PowerNorm(Normalize):
         Determines the behavior for mapping values outside the range
         ``[vmin, vmax]``.
 
-        If clipping is off, values outside the range ``[vmin, vmax]`` are also
-        transformed by the power function, resulting in values outside ``[0, 1]``.
-        This behavior is usually desirable, as colormaps can mark these *under*
-        and *over* values with specific colors.
+        If clipping is off, values above *vmax* are transformed by the power
+        function, resulting in values above 1, and values below *vmin* are linearly
+        transformed resulting in values below 0. This behavior is usually desirable, as
+        colormaps can mark these *under* and *over* values with specific colors.
 
         If clipping is on, values below *vmin* are mapped to 0 and values above
         *vmax* are mapped to 1. Such values become indistinguishable from
@@ -1969,6 +1969,8 @@ class PowerNorm(Normalize):
     .. math::
 
         \left ( \frac{x - v_{min}}{v_{max}  - v_{min}} \right )^{\gamma}
+
+    For input values below *vmin*, gamma is set to one.
     """
     def __init__(self, gamma, vmin=None, vmax=None, clip=False):
         super().__init__(vmin, vmax, clip)
@@ -1994,9 +1996,8 @@ def __call__(self, value, clip=None):
                                      mask=mask)
             resdat = result.data
             resdat -= vmin
-            resdat[resdat < 0] = 0
-            np.power(resdat, gamma, resdat)
-            resdat /= (vmax - vmin) ** gamma
+            resdat /= (vmax - vmin)
+            resdat[resdat > 0] = np.power(resdat[resdat > 0], gamma)
 
             result = np.ma.array(resdat, mask=result.mask, copy=False)
         if is_scalar:
@@ -2006,14 +2007,21 @@ def __call__(self, value, clip=None):
     def inverse(self, value):
         if not self.scaled():
             raise ValueError("Not invertible until scaled")
+
+        result, is_scalar = self.process_value(value)
+
         gamma = self.gamma
         vmin, vmax = self.vmin, self.vmax
 
-        if np.iterable(value):
-            val = np.ma.asarray(value)
-            return np.ma.power(val, 1. / gamma) * (vmax - vmin) + vmin
-        else:
-            return pow(value, 1. / gamma) * (vmax - vmin) + vmin
+        resdat = result.data
+        resdat[resdat > 0] = np.power(resdat[resdat > 0], 1 / gamma)
+        resdat *= (vmax - vmin)
+        resdat += vmin
+
+        result = np.ma.array(resdat, mask=result.mask, copy=False)
+        if is_scalar:
+            result = result[0]
+        return result
 
 
 class BoundaryNorm(Normalize):
 
@@ -552,10 +552,10 @@ def test_colorbar_lognorm_extension(extend):
 
 def test_colorbar_powernorm_extension():
     # Test that colorbar with powernorm is extended correctly
-    f, ax = plt.subplots()
-    cb = Colorbar(ax, norm=PowerNorm(gamma=0.5, vmin=0.0, vmax=1.0),
-                  orientation='vertical', extend='both')
-    assert cb._values[0] >= 0.0
+    # Just a smoke test that adding the colorbar doesn't raise an error or warning
+    fig, ax = plt.subplots()
+    Colorbar(ax, norm=PowerNorm(gamma=0.5, vmin=0.0, vmax=1.0),
+             orientation='vertical', extend='both')
 
 
 def test_colorbar_axes_kw():
 
@@ -555,12 +555,16 @@ def test_PowerNorm():
     assert_array_almost_equal(norm(a), pnorm(a))
 
     a = np.array([-0.5, 0, 2, 4, 8], dtype=float)
-    expected = [0, 0, 1/16, 1/4, 1]
+    expected = [-1/16, 0, 1/16, 1/4, 1]
     pnorm = mcolors.PowerNorm(2, vmin=0, vmax=8)
     assert_array_almost_equal(pnorm(a), expected)
     assert pnorm(a[0]) == expected[0]
     assert pnorm(a[2]) == expected[2]
-    assert_array_almost_equal(a[1:], pnorm.inverse(pnorm(a))[1:])
+    # Check inverse
+    a_roundtrip = pnorm.inverse(pnorm(a))
+    assert_array_almost_equal(a, a_roundtrip)
+    # PowerNorm inverse adds a mask, so check that is correct too
+    assert_array_equal(a_roundtrip.mask, np.zeros(a.shape, dtype=bool))
 
     # Clip = True
     a = np.array([-0.5, 0, 1, 8, 16], dtype=float)
@@ -591,6 +595,15 @@ def test_PowerNorm_translation_invariance():
     assert_array_almost_equal(pnorm(a - 2), expected)
 
 
+def test_powernorm_cbar_limits():
+    fig, ax = plt.subplots()
+    vmin, vmax = 300, 1000
+    data = np.arange(10*10).reshape(10, 10) + vmin
+    im = ax.imshow(data, norm=mcolors.PowerNorm(gamma=0.2, vmin=vmin, vmax=vmax))
+    cbar = fig.colorbar(im)
+    assert cbar.ax.get_ylim() == (vmin, vmax)
+
+
 def test_Normalize():
     norm = mcolors.Normalize()
     vals = np.arange(-10, 10, 1, dtype=float)