function_percentile_plot

Plot any function curve across x-values, grouped by a variable (e.g., scale), with percentile annotations and optional formula text.

---

📥 Arguments

Name	Type	Required	Description
df	pd.DataFrame	✅	DataFrame containing ‘x’, ‘y’, and ‘group’ columns.
xlabel	str	❌	Label for the x-axis. Default: ‘X’.
ylabel	str	❌	Label for the y-axis. Default: ‘Y’.
title	str	❌	Title for the plot. Default: ‘Function Curve with Percentile Labels’.
legend_title	str	❌	Title of the legend. Default: None.
func_loc	str	❌	The location of the function text (if not None). Default: ‘top left’.
legend_loc	str	❌	The location of the legend (if not None). Default: ‘lower right’.
cmap	str	❌	The color map of the lines. Default: ‘viridis’.
hline_at	float	❌	Draw a horizontal line at this y-value. Default: None.
function_str	str	❌	Optional LaTeX string to annotate function formula.
function_str_coords	Tuple[float, float]	❌	Coordinates for the function string annotation in Axes coordinates. Default: (0.05, 0.96).
function_font_size	int	❌	Font size for function string annotation. Default: 13.
numbers_font_size	int	❌	Font size for percentile number annotations. Default: 11.
figsize	tuple	❌	Figure size. Default: (10, 6).
min_label_val	float	❌	Minimum y-value to annotate labels. Default: -inf.
max_label_val	float	❌	Maximum y-value to annotate labels. Default: inf.
percentiles	List[int]	❌	Percentiles to compute for labels. Default: [0,10,20,…,90,95].
save	str	❌	Base filename to save PNG and PDF if provided.

---

📦 Example Output

Click to show example code

import numpy as np
from matplotlib import pyplot as plt
import pandas as pd
from swizz import plot, set_style

# Example function
transform_full = lambda r, c, s: ((np.tanh((r + c) / s) + 1) / (1 + np.tanh(c / s))) * 2 - 1
transform = lambda r, s: transform_full(r, 0, s)

# Generate Data
failed_reward, max_reward = -1500, 0
scores = np.linspace(failed_reward, max_reward, 500)

reward_scales = [500, 750, 1000, 1500, 2000, 3000, 4000, 5000, 7500]

data = []
for scale in reward_scales:
    transformed_scores = transform(scores, scale)
    for x, y in zip(scores, transformed_scores):
        data.append({'x': x, 'y': y, 'group': f'scale={scale}'})

df = pd.DataFrame(data)

# Call the plot method
fig, ax = plot("function_percentile_plot",
    df,
    xlabel="Original Reward",
    ylabel="Transformed Reward",
    title="Transformed Rewards Across Scales",
    cmap="viridis",
    legend_title="Scale",
    min_label_val=-0.98,
    max_label_val=0.8,
    function_str=r'$r_{new} = \frac{\tanh\left(\frac{r + c}{s}\right) + 1}{1 + \tanh\left(\frac{c}{s}\right)} \cdot 2 - 1$',
)
plt.show()