Move a matplotlib Axes from one figure to another.

move_axes.py

  import matplotlib.pyplot as plt

  def move_axes(ax, fig, subplot_spec=111):
      """Move an Axes object from a figure to a new pyplot managed Figure in
      the specified subplot."""

      # get a reference to the old figure context so we can release it
      old_fig = ax.figure

      # remove the Axes from it's original Figure context
      ax.remove()

      # set the pointer from the Axes to the new figure
      ax.figure = fig

      # add the Axes to the registry of axes for the figure
      fig.axes.append(ax)
      # twice, I don't know why...
      fig.add_axes(ax)

      # then to actually show the Axes in the new figure we have to make
      # a subplot with the positions etc for the Axes to go, so make a
      # subplot which will have a dummy Axes
      dummy_ax = fig.add_subplot(subplot_spec)

      # then copy the relevant data from the dummy to the ax
      ax.set_position(dummy_ax.get_position())

      # then remove the dummy
      dummy_ax.remove()

      # close the figure the original axis was bound to
      plt.close(old_fig)

Re: Line 27, I just created a new figure to move all axes onto so that all subplots have the same weird offsets if any

e.g.

all_axes = []
for thing in things:
    axes = make_complicated_plot()
    all_axes.append(axes)

fig = figure()
old_figs = []
for i_row, axes in enumerate(all_axes):
    nrows = len(all_axes)
    ncols = len(axes[0])
    for i_col, ax in enumerate(axes[0]):
        old_fig = move_axes(ax, fig, (nrows, ncols, 1 + i_col + i_row * ncols))
        if i_col == 0:
            old_figs.append(old_fig)

for old_fig in old_figs:
    plt.close(old_fig)

with a couple small tweaks in my fork of this gist

Matplotlib subfigures are also cool for stuff like this, but I found them still a little janky.

@digitalsignalperson What does make_complicated_plot() return? It seems like it should be a matplotlib.axes.Axes, but then you compute len(axes[0]), so the return of make_complicated_plot() must then be subscriptable.

It was using a pandas.DataFrame.plot() return value of a numpy array of Axes

more details:

figsize = (23, 11.5)
nrows = 4
ncols = 7

all_axes = []
for category in categories:
    # group/filter a dataframe based on the category. assume it is df here
    axes = df.plot(kind='pie', subplots=True, layout=(nrows, ncols), figsize=figsize, legend=False, ylabel='', title='example')
    all_axes.append((category, axes))

fig = figure(figsize=figsize)
# subplots_adjust(...) as needed

old_figs = []
for i_row, (cat, axes) in enumerate(all_axes):
    for i_col, ax in enumerate(axes[0]):
        old_fig = move_axes(ax, fig, (nrows, ncols, 1 + i_col + i_row * 7))
        if i_col == 0:
            old_figs.append(old_fig)
            ax.text(-0.25, 0.5, cat, transform=ax.transAxes, rotation=90, ha='center', va='center')
for old_fig in old_figs:
    plt.close(old_fig)

sorry it's not a complete example, but hopefully it works if you play with it

also this is using my fork if it was unclear https://gist.github.com/digitalsignalperson/546e80ae1965b83df0a82ba12ae8aac7

Right, I see. Thanks!