#!/usr/bin/env python3 """ Select 10 diverse clusters for Round 2 testing. Criteria: - 4 from body_is_weird - 3 from power_works - 3 from what_actually_happened Mix: - At least 3 clusters that produced BLAND titles (high score but generic) - At least 3 clusters that produced DECENT titles - At least 2 smaller clusters (under 20 posts) """ import json import os SANDBOX_DIR = os.path.dirname(os.path.abspath(__file__)) OUTPUT_DIR = os.path.join(os.path.dirname(SANDBOX_DIR), 'output') # Markers of generic titles GENERIC_MARKERS = [ "Here's ", "How to ", "A Framework", "Decision Tree", "What You Need", "Everything You", "A Guide", "5 ", "7 ", "10 ", " Steps", " Ways to" ] def is_generic_title(title): """Check if a title is generic/theme-based.""" title_str = str(title) return any(m in title_str for m in GENERIC_MARKERS) def load_channel_data(channel_id): """Load cluster data and opportunities for a channel.""" # Load clusters cluster_path = os.path.join(SANDBOX_DIR, f'{channel_id}_clusters.json') with open(cluster_path) as f: cluster_data = json.load(f) # Load opportunities (existing pipeline output) opp_path = os.path.join(OUTPUT_DIR, f'analysis_opportunities_{channel_id}.json') try: with open(opp_path) as f: opp_data = json.load(f) opportunities = {o.get('cluster_id'): o for o in opp_data.get('opportunities', [])} except FileNotFoundError: opportunities = {} return cluster_data, opportunities def analyze_cluster(cluster, opportunity=None): """Analyze a cluster for selection criteria.""" topics = cluster.get('topics', []) posts = cluster.get('reddit_posts', []) cluster_id = cluster.get('cluster_id') topic_count = cluster.get('topic_count', len(topics)) # Count posts with substantial selftext posts_with_content = sum(1 for p in posts if len(p.get('selftext', '')) > 100) # Get unique subreddits subreddits = list(set(p.get('subreddit', '') for p in posts))[:5] # Get sample topics sample_topics = [t.get('topic_name', '')[:50] for t in topics[:3]] result = { 'cluster_id': cluster_id, 'topic_count': topic_count, 'post_count': len(posts), 'posts_with_content': posts_with_content, 'subreddits': subreddits, 'sample_topics': sample_topics, '_cluster': cluster # Keep reference for extraction } # Add pipeline output info if available if opportunity: title = opportunity.get('suggested_title', '') score = opportunity.get('weighted_score', 0) verdict = opportunity.get('verdict', '') result['pipeline_title'] = str(title)[:80] result['pipeline_score'] = score result['pipeline_verdict'] = verdict result['is_generic'] = is_generic_title(title) return result def select_clusters_for_channel(channel_id, count, clusters, opportunities, need_generic=0, need_decent=0, need_small=0): """Select diverse clusters from a channel.""" # Analyze all clusters analyzed = [] for cluster in clusters: cluster_id = cluster.get('cluster_id') opp = opportunities.get(cluster_id) if opp: # Only consider clusters that have pipeline output analysis = analyze_cluster(cluster, opp) if analysis['posts_with_content'] >= 3: # Minimum content threshold analyzed.append(analysis) # Sort by pipeline score (we want interesting clusters) analyzed.sort(key=lambda x: x.get('pipeline_score', 0), reverse=True) print(f"\n{channel_id}: {len(analyzed)} clusters with pipeline output and content") # Categorize generic = [c for c in analyzed if c.get('is_generic', False)] decent = [c for c in analyzed if not c.get('is_generic', True)] small = [c for c in analyzed if c['topic_count'] < 20] print(f" Generic titles: {len(generic)}, Decent titles: {len(decent)}, Small clusters: {len(small)}") # Select with diversity selected = [] seen_ids = set() # First, get required generic titles for c in generic[:need_generic]: if c['cluster_id'] not in seen_ids: selected.append(c) seen_ids.add(c['cluster_id']) # Then, get required decent titles for c in decent[:need_decent]: if c['cluster_id'] not in seen_ids: selected.append(c) seen_ids.add(c['cluster_id']) # Then, prioritize small clusters if needed for c in small[:need_small]: if c['cluster_id'] not in seen_ids: selected.append(c) seen_ids.add(c['cluster_id']) # Fill remaining slots with highest scored for c in analyzed: if len(selected) >= count: break if c['cluster_id'] not in seen_ids: selected.append(c) seen_ids.add(c['cluster_id']) return selected[:count] def extract_test_data(selected_clusters, channel_id): """Extract full test data for selected clusters.""" test_data = [] for i, metrics in enumerate(selected_clusters): cluster = metrics['_cluster'] # Get all raw posts with their full content raw_posts = [] for post in cluster.get('reddit_posts', []): raw_posts.append({ 'title': post.get('title', ''), 'selftext': post.get('selftext', ''), 'subreddit': post.get('subreddit', ''), 'num_comments': post.get('num_comments', 0), 'score': post.get('score', 0), 'url': post.get('url', '') }) # Get topics topics = [] for t in cluster.get('topics', []): topics.append({ 'topic_name': t.get('topic_name', ''), 'source': t.get('source', ''), 'subreddit': t.get('subreddit', '') }) test_cluster = { 'test_id': f"{channel_id}_cluster_{i+1}", 'cluster_id': metrics['cluster_id'], 'channel': channel_id, 'topic_count': metrics['topic_count'], 'posts_with_content': metrics['posts_with_content'], 'sample_topics': metrics['sample_topics'], 'subreddits': metrics['subreddits'], # Pipeline output for comparison 'pipeline_title': metrics.get('pipeline_title', 'N/A'), 'pipeline_score': metrics.get('pipeline_score', 0), 'pipeline_verdict': metrics.get('pipeline_verdict', 'N/A'), 'is_generic_baseline': metrics.get('is_generic', False), # Raw data 'topics': topics, 'raw_posts': raw_posts } test_data.append(test_cluster) return test_data def main(): print("=" * 70) print("ROUND 2 CLUSTER SELECTION") print("=" * 70) channels = { 'body_is_weird': {'count': 4, 'need_generic': 2, 'need_decent': 1, 'need_small': 1}, 'power_works': {'count': 3, 'need_generic': 1, 'need_decent': 1, 'need_small': 0}, 'what_actually_happened': {'count': 3, 'need_generic': 0, 'need_decent': 1, 'need_small': 1} } all_test_clusters = [] for channel_id, params in channels.items(): # Load data cluster_data, opportunities = load_channel_data(channel_id) clusters = cluster_data.get('clusters', []) # Select clusters selected = select_clusters_for_channel( channel_id, params['count'], clusters, opportunities, params['need_generic'], params['need_decent'], params['need_small'] ) print(f"\nSelected {len(selected)} clusters from {channel_id}:") for c in selected: size_label = "SMALL" if c['topic_count'] < 20 else "MED" if c['topic_count'] < 50 else "LARGE" generic_label = "GENERIC" if c.get('is_generic') else "DECENT" print(f" [{c['cluster_id']}] {size_label} {generic_label} ({c['pipeline_score']:.1f}) \"{c['pipeline_title'][:50]}...\"") # Extract full test data test_data = extract_test_data(selected, channel_id) all_test_clusters.extend(test_data) # Save combined test data output_path = os.path.join(SANDBOX_DIR, 'round2_clusters.json') with open(output_path, 'w') as f: json.dump({ 'round': 2, 'total_clusters': len(all_test_clusters), 'channels': list(channels.keys()), 'clusters': all_test_clusters }, f, indent=2) print(f"\n{'='*70}") print(f"Saved {len(all_test_clusters)} test clusters to {output_path}") print("=" * 70) # Summary total_posts = sum(len(c['raw_posts']) for c in all_test_clusters) total_with_content = sum(c['posts_with_content'] for c in all_test_clusters) generic_count = sum(1 for c in all_test_clusters if c.get('is_generic_baseline')) small_count = sum(1 for c in all_test_clusters if c['topic_count'] < 20) print(f"\nSummary:") print(f" Total raw posts: {total_posts}") print(f" Posts with content: {total_with_content}") print(f" Generic baseline titles: {generic_count}") print(f" Small clusters (<20): {small_count}") if __name__ == "__main__": main()