plots.py

import matplotlib.pyplot as plt
import numpy as np
from inference import cond_ent
import os

# Plots posterior probabilities and conditional entropy per query.
def plot_query_results(results):
    queries = [r["query"] for r in results]
    posteriors = [r["posterior"] for r in results]
    entropies = [r["entropy"] for r in results]
    os.makedirs("results/plots", exist_ok=True)
    x = range(len(queries))

    plt.figure()
    plt.bar(x, posteriors)
    plt.xticks(x, queries, rotation=45, ha="right")
    plt.ylabel("Posterior Probability")
    plt.title("Posterior Probability per Query")
    plt.tight_layout()
    plt.savefig("results/plots/posterior_per_query.png", dpi=300)
    plt.show()

    plt.figure()
    plt.bar(x, entropies)
    plt.xticks(x, queries, rotation=45, ha="right")
    plt.ylabel("Conditional Entropy (bits)")
    plt.title("Conditional Entropy per Query")
    plt.tight_layout()
    plt.savefig("results/plots/cond_entropy_per_query.png", dpi=300)
    plt.show()

#  Visualizes posterior probability and entropy as a function of Bloom filter false positive rate.
def plot_bloom_metrics(fpr_range, prior=0.1):
    os.makedirs("results/plots", exist_ok=True)
    posteriors = [prior / (prior + f * (1 - prior)) for f in fpr_range]
    entropies = [cond_ent(p) for p in posteriors]

    fig, ax1 = plt.subplots(figsize=(10, 6))
    ax1.set_xlabel("Bloom Filter FPR")
    ax1.set_ylabel("Posterior Probability")
    ax1.plot(fpr_range, posteriors)
    ax1.grid(True)

    ax2 = ax1.twinx()
    ax2.set_ylabel("Conditional Entropy (bits)")
    ax2.plot(fpr_range, entropies, linestyle="--")

    plt.title(f"Bloom Filter Trustworthiness (Prior={prior})")
    plt.tight_layout()
    plt.savefig("results/plots/bloom_filter_trustworth.png", dpi=300)
    plt.show()

def plot_mi_threshold_analysis(thresholds, fprs, entropies):
    os.makedirs("results/experiments/plots", exist_ok=True)

    # FPR vs MI threshold
    plt.figure()
    plt.plot(thresholds, fprs, marker='o')
    plt.xscale('log')
    plt.xlabel("Mutual Information Threshold")
    plt.ylabel("Empirical False Positive Rate")
    plt.title("Effect of MI Threshold on Empirical FPR")
    plt.grid(True)
    plt.tight_layout()
    plt.savefig("results/plots/mi_threshold_vs_fpr.png", dpi=300)
    plt.show()

    # Entropy vs MI threshold
    plt.figure()
    plt.plot(thresholds, entropies, marker='o', linestyle='--')
    plt.xscale('log')
    plt.xlabel("Mutual Information Threshold")
    plt.ylabel("Average Conditional Entropy (bits)")
    plt.title("Effect of MI Threshold on Classification Uncertainty")
    plt.grid(True)
    plt.tight_layout()
    plt.savefig("results/plots/mi_threshold_vs_entropy.png", dpi=300)
    plt.show()

    # Trade-off plot (examiner favorite)
    plt.figure()
    plt.scatter(fprs, entropies)
    for i, t in enumerate(thresholds):
        plt.annotate(f"MI={t}", (fprs[i], entropies[i]))
    plt.xlabel("Empirical False Positive Rate")
    plt.ylabel("Average Conditional Entropy (bits)")
    plt.title("Trade-off Between False Positives and Uncertainty")
    plt.grid(True)
    plt.tight_layout()
    plt.savefig("results/plots/fpr_vs_entropy_tradeoff.png", dpi=300)
    plt.show()