Oct 25, 2023 · Oct 23, 2023 · Oct 25, 2023
diff --git a/doc/api/next_api_changes/behavior/27179-KS.rst b/doc/api/next_api_changes/behavior/27179-KS.rst
 Default behavior of ``hexbin`` with *C* provided requires at least 1 point
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 The behavior changed in 3.8.0 to be inclusive of *mincnt*. However that resulted in
 errors or warnings with some reduction functions, so now the default is to require at
 least 1 point to call the reduction function. This effectively restores the default
 behavior to match that of Matplotlib 3.7 and before.
diff --git a/doc/api/prev_api_changes/api_changes_3.8.0/behaviour.rst b/doc/api/prev_api_changes/api_changes_3.8.0/behaviour.rst
 The :rc:`ps.papertype` rcParam can now be set to ``'figure'``, which will use
 a paper size that corresponds exactly with the size of the figure that is being
 saved.

 ``hexbin`` *mincnt* parameter made consistently inclusive
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 Previously, *mincnt* was inclusive with no *C* provided but exclusive when *C* is provided.
 It is now inclusive of *mincnt* in both cases.
diff --git a/lib/matplotlib/axes/_axes.py b/lib/matplotlib/axes/_axes.py
        yscale : {'linear', 'log'}, default: 'linear'
            Use a linear or log10 scale on the vertical axis.

        mincnt : int > 0, default: *None*
            If not *None*, only display cells withmore than *mincnt*
        mincnt : int >= 0, default: *None*
            If not *None*, only display cells withat least *mincnt*
            number of points in the cell.

        marginals : bool, default: *False*
            - `numpy.sum`: integral of the point values
            - `numpy.amax`: value taken from the largest point

            By default will only reduce cells with at least 1 point because some
            reduction functions (such as `numpy.amax`) will error/warn with empty
            input. Changing *mincnt* will adjust the cutoff, and if set to 0 will
            pass empty input to the reduction function.

        data : indexable object, optional
            DATA_PARAMETER_PLACEHOLDER

                else:
                    Cs_at_i2[i2[i]].append(C[i])
            if mincnt is None:
                mincnt =0
                mincnt =1
            accum = np.array(
                [reduce_C_function(acc) if len(acc) >= mincnt else np.nan
                 for Cs_at_i in [Cs_at_i1, Cs_at_i2]
diff --git a/lib/matplotlib/tests/test_axes.py b/lib/matplotlib/tests/test_axes.py
    # From #23922: creating hexbin with log scaling from empty
    # dataset raises ValueError
    ax.hexbin([], [], bins='log')
    # From #27103: np.max errors when handed empty data
    ax.hexbin([], [], C=[], reduce_C_function=np.max)


 def test_hexbin_pickable():
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,7 @@
		Default behavior of ``hexbin`` with C provided requires at least 1 point
		~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

		The behavior changed in 3.8.0 to be inclusive of mincnt. However that resulted in
		errors or warnings with some reduction functions, so now the default is to require at
		least 1 point to call the reduction function. This effectively restores the default
		behavior to match that of Matplotlib 3.7 and before.
Original file line number	Diff line number	Diff line change
Expand Up		@@ -165,3 +165,9 @@ PostScript paper type adds option to use figure size
		The :rc:`ps.papertype` rcParam can now be set to ``'figure'``, which will use
		a paper size that corresponds exactly with the size of the figure that is being
		saved.

		``hexbin`` mincnt parameter made consistently inclusive
		~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

		Previously, mincnt was inclusive with no C provided but exclusive when C is provided.
		It is now inclusive of mincnt in both cases.
Original file line number	Diff line number	Diff line change
Expand Up		@@ -4873,8 +4873,8 @@ def hexbin(self, x, y, C=None, gridsize=100, bins=None,
		yscale : {'linear', 'log'}, default: 'linear'
		Use a linear or log10 scale on the vertical axis.

		mincnt : int > 0, default: None
		If not None, only display cells withmore than mincnt
		mincnt : int >= 0, default: None
		If not None, only display cells withat least mincnt
rcomer marked this conversation as resolved. Show resolvedHide resolved
		number of points in the cell.

		marginals : bool, default: False
Expand DownExpand Up		@@ -4941,6 +4941,11 @@ def reduce_C_function(C: array) -> float
		- `numpy.sum`: integral of the point values
		- `numpy.amax`: value taken from the largest point

		By default will only reduce cells with at least 1 point because some
		reduction functions (such as `numpy.amax`) will error/warn with empty
		input. Changing mincnt will adjust the cutoff, and if set to 0 will
		pass empty input to the reduction function.

		data : indexable object, optional
		DATA_PARAMETER_PLACEHOLDER

Expand DownExpand Up		@@ -5038,7 +5043,7 @@ def reduce_C_function(C: array) -> float
		else:
		Cs_at_i2[i2[i]].append(C[i])
		if mincnt is None:
		mincnt =0
		mincnt =1
		accum = np.array(
		[reduce_C_function(acc) if len(acc) >= mincnt else np.nan
		for Cs_at_i in [Cs_at_i1, Cs_at_i2]
Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -971,6 +971,8 @@ def test_hexbin_empty():
		# From #23922: creating hexbin with log scaling from empty
		# dataset raises ValueError
		ax.hexbin([], [], bins='log')
		# From #27103: np.max errors when handed empty data
		ax.hexbin([], [], C=[], reduce_C_function=np.max)


		def test_hexbin_pickable():
Expand Down