Morris-Frank

Using plots as ticks in PyPlot

Another week another tutorial for something that you never thought about doing. Can we replace the ticks in a PyPlot figure with plots?

Why?

I was building a heatmap showing some result on different signal sources. The signals where different simple curves: sinus, triangle, square and saw-tooth waves. Now for annotating the plot, yes you could write down the names of the signals on the axis and yes that’s probably the ‘correct’ way to do it. But it is ugly and not intuitive if you could also just show the signals:

How?

Unsuprisingly we proceed as follows: We remove all actual ticks from the plot and add correctly small sized axes onto the figure. Adding axes to arbitrary places of a figure is best possible with the inset_axes function provided by the (probably) little known axes_grid1 toolbox in mpl_toolkits (docs).

inset_axes is called as:

inset_axes(ax, width=, height=, bbox_transform=, bbox_to_anchor=, loc=)

We want to position our new axis relative to the axis we position it into (which is ax). To do so we can set bbox_transform to the bounding box of the parent axis: ax.transAxes. bbox_to_anchor will set the anchor/position of the new axis, as we are now setting relative values this can e.g. be (0.05, 1.01) which would be 5% from the left but 1% outside the figure to the bottom. loc sets which point in the new axis is the anchor. For the full list see the documentation, we will need 8 for ’lower center’ and 7 for ‘center right’.

inset_axes will return the new axis which you than can proceed to plot into! So here the full function that will add one tick at a given position on either the upper x axis or the left y axis:

from mpl_toolkits.axes_grid1.inset_locator import inset_axes

def add_plot_tick(
    ax: plt.Axes,
    symbol: str,
    pos: float = 0.5,
    where: str = "x",
    size: float = 0.05
):

    if "x" in where:
        anchor, loc = (pos, 1.01), 8
    else:
        anchor, loc = (-0.025, pos), 7

    _ax = inset_axes(
        ax,
        width=size,
        height=size,
        bbox_transform=ax.transAxes,
        bbox_to_anchor=anchor,
        loc=loc,
    )
    _ax.axison = False

    x = np.linspace(0, τ)

    if "sin" in symbol:
        y = np.sin(x)
    elif "tri" in symbol:
        y = sawtooth(x, width=0.5)
    elif "saw" in symbol:
        y = sawtooth(x, width=1.0)
    elif "sq" in symbol:
        y = square(x)
    else:
        raise ValueError("unknown symbol")

    _ax.plot(x, y, linewidth=3, c="k")

Obviously this code is specific to my use-case of those four signals. One could also just return the axis and then manually plot into it.

To add all four singal ticks we utilize the new function and just loop over the signals. The outer code looks something like:

ax = plt.figure()

ax.imshow(…)
ax.set_axis_off()

pos_tick = np.linspace(0, 1, 2 * 4 + 1)[1::2]
symbols = ["sin", "sq", "tri", "saw"]
for i in range(4):
    add_plot_tick(ax, symbols[i], pos=pos_tick[i], where="x", size=0.25)
    add_plot_tick(ax, symbols[i], pos=pos_tick[-i - 1], where="y", size=0.25)