Skip to content

Instantly share code, notes, and snippets.

@TomasDrozdik
Last active May 18, 2022 23:02
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
Star You must be signed in to star a gist
Save TomasDrozdik/484c5cefbce5b6ab0a1ddd177060af2e to your computer and use it in GitHub Desktop.
Example of interval visualisation on a timeline with seaborn-like interface using matplotlib bars.
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib
import matplotlib.pyplot as plt
def overlapplot(xstart, xend, ycategory, data=None, hue=None, palette=None, **kwargs):
# Load data
start = "start"
end = "end"
cat = "cat"
label_color = "label_color"
df_intervals = pd.DataFrame()
df_intervals[start] = xstart if data is None else data[xstart]
df_intervals[end] = xend if data is None else data[xend]
df_intervals[cat] = ycategory if data is None else data[ycategory]
categories = data[ycategory].unique()
# Figure out coloring and labeling
if hue:
hue = hue if data is None else data[hue]
unique_hue = list(hue.unique())
palette = palette if palette else sns.color_palette(n_colors=len(unique_hue))
df_intervals[label_color] = hue.apply(
lambda x: (x, palette[unique_hue.index(x)])
)
else:
df_overlaps = df_intervals.melt(
id_vars=[cat],
var_name="type",
value_vars=[start, end],
value_name="time",
)
df_overlaps["value"] = np.select(
[df_overlaps["type"] == start, df_overlaps["type"] == end], [1, -1]
)
df_overlaps["overlaps"] = (
df_overlaps.sort_values(by=[cat, "time"])
.groupby(cat)["value"]
.cumsum()
)
df_overlaps[start] = df_overlaps["time"]
df_overlaps[end] = (
df_overlaps.sort_values(by=[cat, "time"])
.groupby(cat)["time"]
.shift(-1, fill_value=None)
)
df_overlaps.dropna(inplace=True)
df_overlaps.drop(
df_overlaps[df_overlaps["overlaps"] == 0].index, inplace=True
)
unique_hue = list(df_overlaps["overlaps"].unique())
palette = palette if palette else sns.color_palette("rocket", len(unique_hue))
df_overlaps[label_color] = df_overlaps["overlaps"].apply(
lambda x: (x, palette[unique_hue.index(x)])
)
# Pass back to intervals since the format is kept
df_intervals = df_overlaps
# Filter nonzero ranges
df_intervals = df_intervals[df_intervals[start] < df_intervals[end]]
# Horizontal bars require xrange tuple in format (x_start, x_length)
df_intervals["xrange"] = df_intervals.apply(
lambda row: (row[start], row[end] - row[start]), axis=1
)
yheight = 1
yheight_shrink = yheight * 0.2
ax = plt.gca(**kwargs)
for label, color in df_intervals[label_color].unique():
for category_idx, category in enumerate(categories):
xranges = df_intervals[
(df_intervals[label_color] == (label, color))
& (df_intervals[cat] == category)
]["xrange"]
ax.broken_barh(
xranges,
yrange=(
category_idx * yheight + yheight_shrink,
yheight - yheight_shrink,
),
color=color,
)
# Create legend
legend_elements = [
matplotlib.patches.Patch(facecolor=x[1], label=x[0])
for x in df_intervals[label_color].unique()
]
ax.legend(handles=legend_elements)
ax.set_yticks(
[
(category_idx * yheight) + yheight / 2
for category_idx in range(len(categories))
],
labels=categories,
)
return ax
xstart = "start"
xend = "end"
category = "pair"
df = pd.concat([
pd.DataFrame({
xstart: np.arange(0, 100, 50),
xend: np.arange(10, 110, 50),
category: "A"
}),
pd.DataFrame({
xstart: np.arange(5, 105, 50),
xend: np.arange(15, 115, 50),
category: "A"
}),
])
ax = overlapplot(xstart, xend, category, data=df)
# More overlapping example
n = 10
a = np.random.normal(100, 20, n)
b = np.random.binomial(200, 0.5, n)
df = pd.concat([
pd.DataFrame({
xstart: a,
xend: a + 40,
category: "A",
}, index=np.arange(0, n)),
pd.DataFrame({
xstart: b,
xend: b + 10,
category: "B",
}, index=np.arange(0, n)),
])
df = df[df[xstart] < df[xend]]
overlapplot(xstart, xend, category, df)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment