import logging
import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
logger = logging.getLogger(__name__)
[docs]
def plot_posterior(
sample,
higher_is_better: bool = False,
min_points_per_hexbin: int = 2,
alg_names: list = None,
filename: str = "posterior.eps",
):
"""
Plots the sample from posterior distribution of a Bayesian statistical test.
Parameters:
-----------
data: An (n x 3) array or DataFrame contaning the probabilities.
alg_names: array of strings. Default np.array(['Alg1', 'Alg2'])
Names of the algorithms under evaluation
Return:
-------
Figure
"""
# Initial Checking
if type(sample) == pd.DataFrame:
sample = sample.values
if sample.ndim == 2:
nrow, ncol = sample.shape
if ncol != 3:
raise ValueError("Initialization ERROR. Incorrect number of dimensions in axis 1.")
else:
raise ValueError("Initialization ERROR. Incorrect number of dimensions for sample")
def transform(p):
lambda1, lambda2, lambda3 = p.T
x = 0.1 * lambda1 + 0.5 * lambda2 + 0.9 * lambda3
y = (0.2 * lambda1 + 1.4 * lambda2 + 0.2 * lambda3) / np.sqrt(3)
return np.vstack((x, y)).T
# Initialize figure
fig = plt.figure(figsize=(5, 5), facecolor="white")
ax = fig.add_axes([0, 0, 1, 1])
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
ax.set_axis_off()
# plot text
if not higher_is_better:
if not alg_names:
ax.text(x=0.5, y=1.4 / np.sqrt(3) + 0.005, s="P(rope)", ha="center", va="bottom")
ax.text(x=0.15, y=0.175 / np.sqrt(3) - 0.005, s="P(alg1<alg2)", ha="right", va="top")
ax.text(x=0.85, y=0.175 / np.sqrt(3) - 0.005, s="P(alg1>alg2)", ha="left", va="top")
else:
ax.text(x=0.5, y=1.4 / np.sqrt(3) + 0.005, s="P(rope)", ha="center", va="bottom")
ax.text(x=0.15, y=0.175 / np.sqrt(3) - 0.005, s="P(" + alg_names[0] + ")", ha="right", va="top")
ax.text(x=0.85, y=0.175 / np.sqrt(3) - 0.005, s="P(" + alg_names[1] + ")", ha="left", va="top")
else:
if not alg_names:
ax.text(x=0.5, y=1.4 / np.sqrt(3) + 0.005, s="P(rope)", ha="center", va="bottom")
ax.text(x=0.15, y=0.175 / np.sqrt(3) - 0.005, s="P(alg2<alg1)", ha="right", va="top")
ax.text(x=0.85, y=0.175 / np.sqrt(3) - 0.005, s="P(alg2>alg1)", ha="left", va="top")
else:
ax.text(x=0.5, y=1.4 / np.sqrt(3) + 0.005, s="P(rope)", ha="center", va="bottom")
ax.text(x=0.15, y=0.175 / np.sqrt(3) - 0.005, s="P(" + alg_names[1] + ")", ha="right", va="top")
ax.text(x=0.85, y=0.175 / np.sqrt(3) - 0.005, s="P(" + alg_names[0] + ")", ha="left", va="top")
# Conversion between barycentric and Cartesian coordinates
sample2d = np.zeros((sample.shape[0], 2))
for p in range(sample.shape[0]):
sample2d[p, :] = transform(sample[p, :])
# Plot projected points
ax.hexbin(sample2d[:, 0], sample2d[:, 1], mincnt=min_points_per_hexbin, cmap=plt.cm.plasma)
# Plot triangle
ax.plot([0.095, 0.505], [0.2 / np.sqrt(3), 1.4 / np.sqrt(3)], linewidth=3.0, color="white")
ax.plot([0.505, 0.905], [1.4 / np.sqrt(3), 0.2 / np.sqrt(3)], linewidth=3.0, color="white")
ax.plot([0.09, 0.905], [0.2 / np.sqrt(3), 0.2 / np.sqrt(3)], linewidth=3.0, color="white")
ax.plot([0.1, 0.5], [0.2 / np.sqrt(3), 1.4 / np.sqrt(3)], linewidth=3.0, color="gray")
ax.plot([0.5, 0.9], [1.4 / np.sqrt(3), 0.2 / np.sqrt(3)], linewidth=3.0, color="gray")
ax.plot([0.1, 0.9], [0.2 / np.sqrt(3), 0.2 / np.sqrt(3)], linewidth=3.0, color="gray")
# plot division lines
ax.plot([0.5, 0.5], [0.2 / np.sqrt(3), 0.6 / np.sqrt(3)], linewidth=3.0, color="gray")
ax.plot([0.3, 0.5], [0.8 / np.sqrt(3), 0.6 / np.sqrt(3)], linewidth=3.0, color="gray")
ax.plot([0.5, 0.7], [0.6 / np.sqrt(3), 0.8 / np.sqrt(3)], linewidth=3.0, color="gray")
if filename:
plt.savefig(filename, bbox_inches="tight")
logger.info(f"Figure {filename} saved to file")
plt.show()
