"""Plotting utilities for experiment results.""" import matplotlib.pyplot as plt import json import os from typing import List, Optional, Dict import logging logger = logging.getLogger(__name__) def plot_learning_curve( baseline: List[float], ours: List[float], output_path: Optional[str] = None, title: str = "Learning Curve Comparison", xlabel: str = "Interaction Round", ylabel: str = "Accuracy", preference_weighted: Optional[List[float]] = None ): """Plot learning curve comparing baseline, contrastive, and optionally preference-weighted methods. Args: baseline: List of accuracy scores for baseline method ours: List of accuracy scores for contrastive method output_path: Path to save plot (optional) title: Plot title xlabel: X-axis label ylabel: Y-axis label preference_weighted: Optional list of accuracy scores for preference-weighted method (2025-style) """ rounds = list(range(1, len(baseline) + 1)) plt.figure(figsize=(10, 6)) plt.plot(rounds, baseline, 'b-o', label='Baseline', linewidth=2, markersize=6) plt.plot(rounds, ours, 'r-s', label='Contrastive', linewidth=2, markersize=6) if preference_weighted is not None and len(preference_weighted) == len(baseline): plt.plot(rounds, preference_weighted, 'g-^', label='Preference-Weighted (2025)', linewidth=2, markersize=6) plt.xlabel(xlabel, fontsize=12) plt.ylabel(ylabel, fontsize=12) plt.title(title, fontsize=14, fontweight='bold') plt.legend(fontsize=11) plt.grid(True, alpha=0.3) plt.ylim([0, 1]) plt.xlim([1, len(baseline)]) # Add value annotations # for i, (b, o) in enumerate(zip(baseline, ours)): # if i == len(baseline) - 1: # Only annotate last point # plt.annotate(f'{b:.3f}', (i+1, b), textcoords="offset points", # xytext=(0,10), ha='center', fontsize=9) # plt.annotate(f'{o:.3f}', (i+1, o), textcoords="offset points", # xytext=(0,-15), ha='center', fontsize=9) plt.tight_layout() if output_path: plt.savefig(output_path, dpi=300, bbox_inches='tight') logger.info(f"Plot saved to {output_path}") else: plt.show() plt.close() def plot_all_metrics_func( baseline_metrics: Dict[str, List[float]], contrastive_metrics: Dict[str, List[float]], output_path: Optional[str] = None, preference_weighted_metrics: Optional[Dict[str, List[float]]] = None ): """Plot all metrics (accuracy, precision, recall, F1) in subplots. Args: baseline_metrics: Dictionary with 'accuracy', 'precision', 'recall', 'f1' lists contrastive_metrics: Dictionary with 'accuracy', 'precision', 'recall', 'f1' lists output_path: Path to save plot (optional) preference_weighted_metrics: Optional third method metrics (2025-style) """ rounds = list(range(1, len(baseline_metrics['accuracy']) + 1)) fig, axes = plt.subplots(2, 2, figsize=(14, 10)) fig.suptitle('Evaluation Metrics Comparison', fontsize=16, fontweight='bold') metrics = [ ('accuracy', 'Accuracy', axes[0, 0]), ('precision', 'Precision', axes[0, 1]), ('recall', 'Recall', axes[1, 0]), ('f1', 'F1 Score', axes[1, 1]) ] for metric_key, metric_name, ax in metrics: baseline = baseline_metrics[metric_key] contrastive = contrastive_metrics[metric_key] ax.plot(rounds, baseline, 'b-o', label='Baseline', linewidth=2, markersize=5) ax.plot(rounds, contrastive, 'r-s', label='Contrastive', linewidth=2, markersize=5) if preference_weighted_metrics is not None and metric_key in preference_weighted_metrics: pw = preference_weighted_metrics[metric_key] if len(pw) == len(baseline): ax.plot(rounds, pw, 'g-^', label='Preference-Weighted (2025)', linewidth=2, markersize=5) ax.set_xlabel('Interaction Round', fontsize=10) ax.set_ylabel(metric_name, fontsize=10) ax.set_title(metric_name, fontsize=12, fontweight='bold') ax.legend(fontsize=9) ax.grid(True, alpha=0.3) ax.set_ylim([0, 1]) ax.set_xlim([1, len(baseline)]) # Add final value annotations if len(baseline) > 0: ax.annotate(f'{baseline[-1]:.3f}', (len(baseline), baseline[-1]), textcoords="offset points", xytext=(5, 0), ha='left', fontsize=8) ax.annotate(f'{contrastive[-1]:.3f}', (len(contrastive), contrastive[-1]), textcoords="offset points", xytext=(5, -12), ha='left', fontsize=8) plt.tight_layout() if output_path: plt.savefig(output_path, dpi=300, bbox_inches='tight') logger.info(f"All metrics plot saved to {output_path}") else: plt.show() plt.close() def plot_from_json( baseline_path: str, contrastive_path: str, output_path: Optional[str] = None, plot_all_metrics: bool = False, preference_weighted_path: Optional[str] = None ): """Load results from JSON files and plot. Args: baseline_path: Path to baseline metrics JSON file contrastive_path: Path to contrastive metrics JSON file output_path: Path to save plot (optional) plot_all_metrics: If True, plot all metrics in subplots; if False, plot only accuracy preference_weighted_path: Optional path to preference-weighted metrics JSON (2025-style third method) """ with open(baseline_path, 'r') as f: baseline_data = json.load(f) with open(contrastive_path, 'r') as f: contrastive_data = json.load(f) # Handle both old format (list) and new format (dict) if isinstance(baseline_data, list): baseline_metrics = {'accuracy': baseline_data} contrastive_metrics = {'accuracy': contrastive_data} else: baseline_metrics = baseline_data contrastive_metrics = contrastive_data pw_metrics = None if preference_weighted_path and os.path.isfile(preference_weighted_path): with open(preference_weighted_path, 'r') as f: pw_data = json.load(f) pw_metrics = pw_data if isinstance(pw_data, dict) else {'accuracy': pw_data} if plot_all_metrics and 'precision' in baseline_metrics: from experiment.plotter import plot_all_metrics_func plot_all_metrics_func(baseline_metrics, contrastive_metrics, output_path, preference_weighted_metrics=pw_metrics) else: pw_acc = pw_metrics['accuracy'] if pw_metrics else None plot_learning_curve( baseline_metrics['accuracy'], contrastive_metrics['accuracy'], output_path=output_path, preference_weighted=pw_acc ) if __name__ == "__main__": import argparse parser = argparse.ArgumentParser(description="Plot learning curves") parser.add_argument("--baseline", type=str, default="results/metrics_baseline.json", help="Path to baseline metrics JSON") parser.add_argument("--contrastive", type=str, default="results/metrics_contrastive.json", help="Path to contrastive metrics JSON") parser.add_argument("--preference-weighted", type=str, default=None, help="Path to preference-weighted metrics JSON (optional third method)") parser.add_argument("--output", type=str, default="results/learning_curve.png", help="Output path for plot") parser.add_argument("--accuracy-only", action="store_true", help="Only plot accuracy (single figure); default when not using --all-metrics") parser.add_argument("--all-metrics", action="store_true", help="Plot all metrics (accuracy, precision, recall, F1) in 2x2 subplots") args = parser.parse_args() logging.basicConfig( level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s' ) # 仅当未指定 --all-metrics 时画单独 accuracy 图；--accuracy-only 则强制只画 accuracy plot_all = args.all_metrics and not args.accuracy_only plot_from_json( args.baseline, args.contrastive, args.output, plot_all_metrics=plot_all, preference_weighted_path=args.preference_weighted )