Problem with passing a PathCollection object as the objects argument

[the-pawel-wojcik]

Hello,

A very helpful tool! All works great when I work with text objects alone. I run into troubles after asking for a bit more complex task.

I would like to use adjust_text to avoid a set of lines that I draw using vlines. When I take the output and turn it into PathCollection object I run into errors. It could be that I just don't know how to use PathCollection.

I am using matplotlib 3.9.2 and python 3.12.3.

An example:

import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
from adjustText import adjust_text

# prepare input
npeaks = 10
xs = np.linspace(0, 1, npeaks)
rng = np.random.default_rng(seed=1)
heights = rng.random(size=npeaks)

# plot stems
fig, ax = plt.subplots(figsize=(3, 3))
vlines: mpl.collections.LineCollection = ax.vlines(xs, heights, 0.0)

# add texts
texts = list()
for x, y in zip(xs, heights):
    text = ax.text(x, y/2, "note", ha='center', va='center')
    texts += [text]

# turn the LineCollection into a PathCollection
path_list: list[mpl.path.Path] = vlines.get_paths()
print(f"{path_list[0]=}")
pathcoll = mpl.collections.PathCollection(path_list)
print(f"{pathcoll=}")

adjust_text(texts, objects=pathcoll)

plt.show()

Thanks
Pawel

[BBC-Esq]

The problem is twofold:

1. PathCollection created manually lacks required attributes - The get_bboxes_pathcollection function expects a scatter-plot-style PathCollection with _offsets set up properly. Your manually created PathCollection doesn't have this.

2. Lines have zero width - Vertical lines have bounding boxes with zero width, making collision detection unreliable.

Solution: Create Bbox objects with padding

import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
from adjustText import adjust_text

# prepare input
npeaks = 10
xs = np.linspace(0, 1, npeaks)
rng = np.random.default_rng(seed=1)
heights = rng.random(size=npeaks)

# plot stems
fig, ax = plt.subplots(figsize=(3, 3))
vlines: mpl.collections.LineCollection = ax.vlines(xs, 0.0, heights)

# add texts
texts = []
for x, y in zip(xs, heights):
    text = ax.text(x, y / 2, "note", ha="center", va="center")
    texts.append(text)

# Draw first so transforms are valid
fig.canvas.draw()

# Create Bbox objects from the vlines with padding (since lines have zero width)
line_width_padding = 0.01  # in data coordinates
bboxes = []
for x, h in zip(xs, heights):
    bbox_data = mpl.transforms.Bbox([
        [x - line_width_padding, 0],
        [x + line_width_padding, h]
    ])
    # Transform to display coordinates to match text bboxes
    bbox_display = bbox_data.transformed(ax.transData)
    bboxes.append(bbox_display)

adjust_text(texts, objects=bboxes)
plt.show()

Alternative: Just use x and y parameters

Since your lines are vertical at known x positions, you can simply tell adjust_text to avoid those points:

# Create multiple y points along each line for better avoidance
avoid_x = []
avoid_y = []
for x, h in zip(xs, heights):
    for y_point in np.linspace(0, h, 5):  # 5 points per line
        avoid_x.append(x)
        avoid_y.append(y_point)

adjust_text(texts, x=avoid_x, y=avoid_y)

This second approach is simpler and avoids the Bbox coordinate system issues entirely.