src/synthetic/Table1_utils/make_appendix_f1_vs_pout_tikz.py

import csv
from collections import defaultdict
from statistics import mean, stdev
import math


def rename_method(label: str) -> str:
    """Map internal method keys to the short legend labels we want in the paper."""
    mapping = {
        'Spectral': 'Spectral Clustering',
        'Densest': 'Densest Subgraph',
        'QuasiClique': 'QuasiClique',
        'k-core': 'k-core',
        'L-RMC': 'L-RMC (ours)',
    }
    return mapping.get(label, label)


def load_rows(path):
    with open(path, 'r') as f:
        reader = csv.DictReader(f)
        return [row for row in reader]


def aggregate_f1_vs_pout(rows):
    # Group by (Method, p_out) across k and p_in
    group = defaultdict(list)
    for row in rows:
        method = row['Method']
        try:
            pout = float(row['ExternalDensity'])
            f1 = float(row['F1'])
        except (KeyError, ValueError):
            continue
        group[(method, pout)].append(f1)

    summary = defaultdict(dict)  # method -> pout -> (mean, ci95, n)
    for (method, pout), vals in group.items():
        n = len(vals)
        mu = mean(vals)
        if n > 1:
            s = stdev(vals)
        else:
            s = 0.0
        ci95 = 1.96 * s / math.sqrt(max(n, 1))
        summary[method][pout] = (mu, ci95, n)
    return summary


def emit_tikz(summary, out_path):
    # Prepare sorted p_outs and methods
    pouts = sorted({p for m in summary for p in summary[m].keys()})
    methods = [
        'L-RMC', 'QuasiClique', 'Spectral', 'Densest', 'k-core'
    ]
    methods = [m for m in methods if m in summary]

    # Colors list for methods
    colors = [
        'blue!80!black', 'orange!80!black', 'green!60!black', 'red!80!black', 'purple!70!black'
    ]

    with open(out_path, 'w') as w:
        w.write('% Auto-generated by make_appendix_f1_vs_pout_tikz.py\n')
        w.write('\\begin{figure}[h]\n')
        w.write('  \\centering\n')
        w.write('  \\begin{tikzpicture}\n')
        w.write('    \\begin{axis}[\n')
        w.write('      width=0.9\\linewidth, height=5cm,\n')
        w.write('      xlabel={$p_{\\text{out}}$}, ylabel={F1},\n')
        w.write('      xmin=0.22, xmax=0.48, ymin=0, ymax=1.02,\n')
        w.write('      grid=both,\n')
        w.write('      legend to name=f1legend,\n')
        w.write('      legend columns=3,\n')
        w.write('      legend cell align=left,\n')
        w.write('      legend image post style={xscale=1.2},\n')
        w.write('      legend style={draw=none, fill=none, font=\\scriptsize, /tikz/every even column/.append style={column sep=0.75em}}\n')
        w.write('    ]\n\n')

        for idx, method in enumerate(methods):
            label = rename_method(method)
            color = colors[idx % len(colors)]

            # Means and 95% CI bands
            coords = []
            upper = []
            lower = []
            for p in pouts:
                if p in summary[method]:
                    mu, ci95, _ = summary[method][p]
                    coords.append((p, mu))
                    upper.append((p, mu + ci95))
                    lower.append((p, mu - ci95))

            name = f"{method.replace('-', '').replace(' ', '')}"

            # Optional series comment for readability
            comment = label
            if method == 'L-RMC':
                comment = 'L-RMC (ours)'
            w.write(f"    % --- {comment}\n")

            # Mean line
            w.write(f"    \\addplot+[mark=o, thick, draw={color}]\n")
            w.write("      coordinates {")
            for (x, y) in coords:
                w.write(f" ({x:.2f},{y:.3f})")
            w.write("};\n")
            w.write(f"    \\addlegendentry{{{label}}}\n")

            # Upper and lower paths for shading
            w.write(f"    \\addplot[name path=upper{name}, draw=none, forget plot]\n")
            w.write("      coordinates {")
            for (x, y) in upper:
                w.write(f" ({x:.2f},{y:.3f})")
            w.write("};\n")
            w.write(f"    \\addplot[name path=lower{name}, draw=none, forget plot]\n")
            w.write("      coordinates {")
            for (x, y) in lower:
                w.write(f" ({x:.2f},{y:.3f})")
            w.write("};\n")

            # Fill between with specified style
            w.write(f"    \\addplot[fill={color}, fill opacity=0.30, draw=none, forget plot]\n")
            w.write(f"      fill between [of=upper{name} and lower{name}];\n\n")

        w.write('    \\end{axis}\n')
        w.write('  \\end{tikzpicture}\n\n')
        w.write('  % Place the short legend collected above\n')
        w.write('  \\pgfplotslegendfromname{f1legend}\n\n')
        w.write('  \\caption{F1 vs $p_{\\text{out}}$ (averaged over $k$ and $p_{\\text{in}}$). '
                'Shaded bands show 95\\% CIs. '
                'The baselines are fully described in Section~\\ref{sec:synthetic-planted}. '
                'Spectral Clustering uses the principal non-trivial eigenvector; '
                'Densest Subgraph is Charikar\'s peeling; '
                'k-core reports the densest component; '
                'QuasiClique optimizes edge density.}\n')
        w.write('\\end{figure}\n')
    print(f"Wrote {out_path}")


if __name__ == '__main__':
    import argparse
    ap = argparse.ArgumentParser()
    ap.add_argument('csv', help='Path to Table1 CSV (per-setting averages)')
    ap.add_argument('--out', default='appendix_f1_vs_pout.tikz', help='Output TikZ file to include')
    args = ap.parse_args()

    rows = load_rows(args.csv)
    summary = aggregate_f1_vs_pout(rows)
    emit_tikz(summary, args.out)