results-and-visualization

name	results-and-visualization
description	Publication-quality visualization patterns using matplotlib for convergence curves, worker trajectories, performance breakdowns, population diversity plots, and BO comparison charts. Use this skill when creating visualizations, plots, figures for the paper, analyzing results JSON files, rendering multi-seed error bands, or designing any figure for academic publication or presentation.

Results and Visualization

Results JSON Schema

All PBT runs output a results JSON with this structure:

{
  "system_info": { "cpu_model": "...", "ram_total_gb": 16.0, ... },
  "experiment_config": { "workload": "oltp", "tier": "core", "seed": 42, ... },
  "generation_history": [
    {
      "generation": 1,
      "best_score": 72.5,
      "mean_score": 45.3,
      "std_score": 12.1,
      "median_score": 44.0,
      "worker_scores": [72.5, 45.3, 38.1, ...],
      "exploit_events": [
        {"source": 0, "target": 3, "source_score": 72.5, "target_score": 38.1}
      ]
    }
  ],
  "best_config": { "shared_buffers": 0.30, "work_mem": 0.04, ... },
  "best_score": 85.3,
  "total_time_seconds": 3600.0
}

Parsing Pattern

import json

with open('results/path/to/results.json') as f:
    data = json.load(f)

gen_history = data['generation_history']
best_scores = [g['best_score'] for g in gen_history]
mean_scores = [g['mean_score'] for g in gen_history]
std_scores = [g['std_score'] for g in gen_history]
worker_scores = [g.get('worker_scores', []) for g in gen_history]
exploit_events = [g.get('exploit_events', []) for g in gen_history]

Approved Plot Types

Plot 1: Convergence Curve

• Running-best score + per-generation best/mean ± std band
• Baseline horizontal line for reference
• Exploit-event markers (vertical dashed lines or scatter points)
• Data source: generation_history → best_score, mean_score, std_score

Plot 2: Wall-Clock Time Comparison (PBT vs BO)

• Score vs elapsed time (seconds)
• Shows PBT's parallelism advantage
• Both methods on same axes for direct comparison

Plot 3: Sample Efficiency Comparison (PBT vs BO)

• Score vs total evaluations (not wall-clock)
• Shows BO's potential sample efficiency advantage
• Total evals = population_size × generations for PBT

Plot 4: Per-Worker Trajectory

• Individual worker scores per generation (thin lines)
• Exploit-event markers showing when workers were replaced
• Best worker highlighted with thick line
• Data source: worker_scores in generation_history

Plot 5: Performance Improvement Breakdown

• Grouped bar chart: latency improvement, throughput improvement, memory savings
• Multi-seed error bars (mean ± std across 5 seeds)
• Relative to baseline (0% = default PG config)

Plot 6: Population Diversity Over Time

• std_score plotted over generations
• Shows convergence behavior — diversity should decrease over time
• Sharp drops indicate successful exploit events

Academic Figure Standards

import matplotlib.pyplot as plt
import matplotlib as mpl

# Recommended style setup
plt.style.use('seaborn-v0_8-paper')
mpl.rcParams.update({
    'font.size': 10,
    'axes.labelsize': 12,
    'axes.titlesize': 14,
    'xtick.labelsize': 10,
    'ytick.labelsize': 10,
    'legend.fontsize': 10,
    'figure.dpi': 300,
    'savefig.dpi': 300,
    'savefig.bbox': 'tight',
})

# Use LaTeX labels if available
try:
    plt.rc('text', usetex=True)
    plt.rc('font', family='serif')
except:
    pass  # Fall back to matplotlib's mathtext

Figure Sizing

Layout	Width	Use Case
Single-column	3.5 in	Conference papers
Double-column	7.0 in	Wide figures, comparison charts
Full-page	7.0 × 9.0 in	Multi-panel figures

Color Palette

Use colorblind-friendly schemes:

COLORS = {
    'pbt': '#2196F3',      # Blue
    'bo': '#FF9800',       # Orange
    'baseline': '#9E9E9E', # Gray
    'best': '#4CAF50',     # Green
    'exploit': '#F44336',  # Red (markers)
}

Save Format

fig.savefig('figure_1.pdf')   # Vector (for paper)
fig.savefig('figure_1.png')   # Raster (for preview)

Multi-Seed Error Band Rendering

import numpy as np

# all_seeds_scores: shape (n_seeds, n_generations)
mean_scores = np.mean(all_seeds_scores, axis=0)
std_scores = np.std(all_seeds_scores, axis=0)
generations = np.arange(1, len(mean_scores) + 1)

fig, ax = plt.subplots(figsize=(7, 4))
ax.fill_between(generations,
    mean_scores - std_scores,
    mean_scores + std_scores,
    alpha=0.3, color=COLORS['pbt'])
ax.plot(generations, mean_scores, linewidth=2,
    color=COLORS['pbt'], label='PBT (mean ± std)')
ax.axhline(y=baseline_score, color=COLORS['baseline'],
    linestyle='--', label='Default PG')
ax.set_xlabel('Generation')
ax.set_ylabel('Performance Score')
ax.legend()