plot_hypothesis_tests.py

import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
import pandas as pd
import subprocess
from io import StringIO
import scipy.stats

matplotlib.use("Agg")
sns.set()
matplotlib.rcParams["text.usetex"] = True
plt.rc("font", family="serif")
sns.set_style("whitegrid")

num_samples = 1000000
output = subprocess.run(["./build/test/MMD", str(num_samples)], capture_output=True)
mmd_df = pd.read_csv(StringIO(output.stdout.decode("utf-8")))
output = subprocess.run(
    ["./build/test/ChiSquared", str(num_samples)], capture_output=True
)
chi_df = pd.read_csv(StringIO(output.stdout.decode("utf-8")))
# Add p value
sig1 = scipy.stats.chi2.ppf(1 - 0.01, df=119)
for r in range(chi_df["Rounds"].min(), chi_df["Rounds"].max() + 1):
    chi_df = chi_df.append(
        {"Algorithm": "$\\alpha=0.01$", "$\chi^2$": sig1, "Rounds": r},
        ignore_index=True,
    )

commbined = mmd_df["Algorithm"].append(chi_df["Algorithm"])
unique = commbined.unique()
palette = dict(zip(unique, sns.color_palette(n_colors=len(unique))))

for n in mmd_df["n"].unique():
    plt.figure(figsize=(6, 4.5))
    algs = mmd_df["Algorithm"].unique()
    sns.lineplot(
        data=mmd_df.loc[mmd_df["n"] == n],
        x="Rounds",
        y="$|\hat{\mathrm{MMD}}^2|$",
        hue="Algorithm",
        style="Algorithm",
        palette=palette,
    )
    plt.yscale("log")
    plt.savefig("MMD_n{}.png".format(n), bbox_inches="tight", dpi=1000)
    plt.clf()

plt.figure(figsize=(6, 4.5))
sns.lineplot(
    data=chi_df,
    x="Rounds",
    y="$\chi^2$",
    hue="Algorithm",
    style="Algorithm",
    palette=palette,
)
plt.yscale("log")
plt.savefig("ChiSquared.png", bbox_inches="tight", dpi=1000)