claude-evolve 1.8.51 → 1.9.0

package/lib/evolve_ideate.py

@@ -0,0 +1,509 @@
+#!/usr/bin/env python3
+"""
+Ideation module for claude-evolve.
+Generates new algorithm ideas using various strategies.
+
+AIDEV-NOTE: This is the Python port of bin/claude-evolve-ideate.
+Includes novelty filtering to prevent near-duplicate ideas.
+"""
+
+import argparse
+import os
+import re
+import shutil
+import sys
+import tempfile
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import List, Optional, Dict, Tuple
+
+# Add lib to path
+SCRIPT_DIR = Path(__file__).parent
+sys.path.insert(0, str(SCRIPT_DIR.parent))
+
+from lib.evolution_csv import EvolutionCSV
+from lib.ai_cli import call_ai, get_git_protection_warning, AIError
+from lib.embedding import check_novelty as check_embedding_novelty, get_embedding, set_cache_file, save_cache
+
+
+@dataclass
+class IdeationConfig:
+    """Configuration for ideation."""
+    csv_path: str
+    evolution_dir: str
+    brief_path: str
+    algorithm_path: str
+
+    # Strategy counts
+    total_ideas: int = 15
+    novel_exploration: int = 3
+    hill_climbing: int = 5
+    structural_mutation: int = 3
+    crossover_hybrid: int = 4
+    num_elites: int = 3
+
+    # Novelty filtering
+    novelty_enabled: bool = True
+    novelty_threshold: float = 0.92
+
+
+@dataclass
+class Idea:
+    """A generated idea."""
+    id: str
+    based_on_id: str
+    description: str
+    strategy: str
+
+
+@dataclass
+class IdeationContext:
+    """Context for ideation strategies."""
+    generation: int
+    top_performers: List[Dict]
+    brief_content: str
+    existing_descriptions: List[str]
+    config: IdeationConfig
+
+
+class IdeationStrategy(ABC):
+    """Base class for ideation strategies."""
+
+    def __init__(self, config: IdeationConfig, csv: EvolutionCSV):
+        self.config = config
+        self.csv = csv
+
+    @property
+    @abstractmethod
+    def name(self) -> str:
+        """Strategy name."""
+        pass
+
+    @abstractmethod
+    def build_prompt(self, context: IdeationContext, ids: List[str], temp_csv_basename: str) -> str:
+        """Build the AI prompt."""
+        pass
+
+    def generate(self, context: IdeationContext, count: int) -> List[Idea]:
+        """Generate ideas using this strategy."""
+        if count <= 0:
+            return []
+
+        print(f"[IDEATE] Running {self.name} strategy for {count} ideas", file=sys.stderr)
+
+        # Get next IDs
+        ids = self.csv.get_next_ids(context.generation, count)
+        print(f"[IDEATE] Using IDs: {', '.join(ids)}", file=sys.stderr)
+
+        # Create temp CSV with stub rows
+        temp_csv = Path(self.config.evolution_dir) / f"temp-csv-{os.getpid()}.csv"
+        shutil.copy(self.config.csv_path, temp_csv)
+
+        # Add stub rows
+        with open(temp_csv, 'a') as f:
+            for id in ids:
+                parent = self._get_default_parent(context)
+                f.write(f'{id},{parent},"[PLACEHOLDER: Replace with algorithmic idea]",,pending\n')
+
+        try:
+            # Build prompt
+            prompt = self.build_prompt(context, ids, temp_csv.name)
+
+            # Call AI
+            output, model = call_ai(prompt, command="ideate", working_dir=self.config.evolution_dir)
+
+            # Parse results from modified CSV
+            ideas = self._parse_results(temp_csv, ids)
+
+            if ideas:
+                # Record model used
+                for idea in ideas:
+                    idea.strategy = f"{self.name} ({model})"
+
+            return ideas
+
+        except AIError as e:
+            print(f"[IDEATE] AI error in {self.name}: {e}", file=sys.stderr)
+            return []
+        finally:
+            temp_csv.unlink(missing_ok=True)
+
+    def _get_default_parent(self, context: IdeationContext) -> str:
+        """Get default parent ID for this strategy."""
+        if context.top_performers:
+            return context.top_performers[0]['id']
+        return ""
+
+    def _parse_results(self, temp_csv: Path, expected_ids: List[str]) -> List[Idea]:
+        """Parse ideas from modified CSV."""
+        ideas = []
+
+        with open(temp_csv) as f:
+            import csv
+            reader = csv.reader(f)
+            for row in reader:
+                if len(row) >= 3:
+                    id = row[0].strip().strip('"')
+                    if id in expected_ids:
+                        based_on = row[1].strip() if len(row) > 1 else ""
+                        description = row[2].strip().strip('"')
+                        # Skip if still placeholder
+                        if "PLACEHOLDER" not in description and description:
+                            ideas.append(Idea(
+                                id=id,
+                                based_on_id=based_on,
+                                description=description,
+                                strategy=self.name
+                            ))
+
+        return ideas
+
+
+class NovelExplorationStrategy(IdeationStrategy):
+    """Generate novel, creative ideas not based on existing algorithms."""
+
+    @property
+    def name(self) -> str:
+        return "novel_exploration"
+
+    def _get_default_parent(self, context: IdeationContext) -> str:
+        return ""  # Novel ideas have no parent
+
+    def build_prompt(self, context: IdeationContext, ids: List[str], temp_csv_basename: str) -> str:
+        return f"""{get_git_protection_warning()}
+
+I need you to use your file editing capabilities to fill in PLACEHOLDER descriptions in the CSV file: {temp_csv_basename}
+
+Current evolution context:
+- Generation: {context.generation}
+- Brief: {context.brief_content[:500]}
+
+CRITICAL TASK:
+The CSV file already contains stub rows with these IDs: {', '.join(ids)}
+Each stub row has a PLACEHOLDER description.
+Your job is to REPLACE each PLACEHOLDER with a real algorithmic idea description.
+
+IMPORTANT FILE READING INSTRUCTIONS:
+Read ONLY the last 20-30 lines of the CSV file to see the placeholder rows.
+DO NOT read the entire file - use offset and limit parameters.
+
+CRITICAL INSTRUCTIONS:
+1. Read ONLY the last 20-30 lines of the CSV to see the placeholder rows
+2. DO NOT ADD OR DELETE ANY ROWS - only EDIT the placeholder descriptions
+3. DO NOT CHANGE THE IDs - they are already correct
+4. Use the Edit tool to replace EACH PLACEHOLDER text with a real algorithmic idea
+5. ALWAYS wrap the description field in double quotes
+6. Each description should be one clear sentence describing a novel algorithmic approach
+7. Focus on creative, ambitious ideas that haven't been tried yet
+
+IMPORTANT: Use your file editing tools (Edit/MultiEdit) to modify the CSV file directly."""
+
+
+class HillClimbingStrategy(IdeationStrategy):
+    """Generate incremental improvements to top performers."""
+
+    @property
+    def name(self) -> str:
+        return "hill_climbing"
+
+    def build_prompt(self, context: IdeationContext, ids: List[str], temp_csv_basename: str) -> str:
+        top_str = "\n".join(
+            f"  {p['id']}: {p['description'][:100]}... (score: {p['performance']})"
+            for p in context.top_performers[:5]
+        )
+        valid_parents = ",".join(p['id'] for p in context.top_performers[:5])
+
+        return f"""{get_git_protection_warning()}
+
+I need you to use your file editing capabilities to fill in PLACEHOLDER descriptions in the CSV file: {temp_csv_basename}
+
+IMPORTANT: You MUST use one of these exact parent IDs: {valid_parents}
+
+Successful algorithms to tune:
+{top_str}
+
+CRITICAL TASK:
+The CSV file already contains stub rows with these IDs: {', '.join(ids)}
+Your job is to REPLACE each PLACEHOLDER with a parameter tuning idea.
+
+INSTRUCTIONS:
+1. Read ONLY the last 20-30 lines of the CSV file
+2. Each idea should be a small parameter adjustment or optimization
+3. Reference which parent you're improving and what specifically you're changing
+4. DO NOT ADD OR DELETE ANY ROWS - only EDIT the placeholder descriptions
+5. ALWAYS wrap descriptions in double quotes
+6. Use the Edit tool to modify the file directly"""
+
+
+class StructuralMutationStrategy(IdeationStrategy):
+    """Generate structural changes to algorithms."""
+
+    @property
+    def name(self) -> str:
+        return "structural_mutation"
+
+    def build_prompt(self, context: IdeationContext, ids: List[str], temp_csv_basename: str) -> str:
+        top_str = "\n".join(
+            f"  {p['id']}: {p['description'][:100]}..."
+            for p in context.top_performers[:5]
+        )
+        valid_parents = ",".join(p['id'] for p in context.top_performers[:5])
+
+        return f"""{get_git_protection_warning()}
+
+I need you to use your file editing capabilities to fill in PLACEHOLDER descriptions in the CSV file: {temp_csv_basename}
+
+IMPORTANT: You MUST use one of these exact parent IDs: {valid_parents}
+
+Top algorithms for structural changes:
+{top_str}
+
+CRITICAL TASK:
+The CSV file already contains stub rows with these IDs: {', '.join(ids)}
+Your job is to REPLACE each PLACEHOLDER with a structural mutation idea.
+
+INSTRUCTIONS:
+1. Read ONLY the last 20-30 lines of the CSV file
+2. Each idea should involve a significant architectural change
+3. Examples: adding new features, changing data flow, combining techniques
+4. DO NOT ADD OR DELETE ANY ROWS - only EDIT the placeholder descriptions
+5. ALWAYS wrap descriptions in double quotes
+6. Use the Edit tool to modify the file directly"""
+
+
+class CrossoverStrategy(IdeationStrategy):
+    """Generate crossover ideas combining multiple algorithms."""
+
+    @property
+    def name(self) -> str:
+        return "crossover"
+
+    def build_prompt(self, context: IdeationContext, ids: List[str], temp_csv_basename: str) -> str:
+        top_str = "\n".join(
+            f"  {p['id']}: {p['description'][:100]}..."
+            for p in context.top_performers[:5]
+        )
+        valid_parents = ",".join(p['id'] for p in context.top_performers[:5])
+
+        return f"""{get_git_protection_warning()}
+
+I need you to use your file editing capabilities to fill in PLACEHOLDER descriptions in the CSV file: {temp_csv_basename}
+
+IMPORTANT: Reference multiple parents from: {valid_parents}
+
+Top algorithms to combine:
+{top_str}
+
+CRITICAL TASK:
+The CSV file already contains stub rows with these IDs: {', '.join(ids)}
+Your job is to REPLACE each PLACEHOLDER with a crossover idea.
+
+INSTRUCTIONS:
+1. Read ONLY the last 20-30 lines of the CSV file
+2. Each idea should combine elements from 2+ top algorithms
+3. In parent_id, list the main parent (use comma-separated for multiple)
+4. Describe how you're combining the approaches
+5. DO NOT ADD OR DELETE ANY ROWS - only EDIT the placeholder descriptions
+6. ALWAYS wrap descriptions in double quotes
+7. Use the Edit tool to modify the file directly"""
+
+
+class Ideator:
+    """Main ideation controller."""
+
+    def __init__(self, config: IdeationConfig):
+        self.config = config
+        self.csv = EvolutionCSV(config.csv_path)
+
+        # Initialize embedding cache for novelty filtering
+        if config.novelty_enabled:
+            cache_path = Path(config.evolution_dir) / "embeddings_cache.json"
+            set_cache_file(str(cache_path))
+            print(f"[IDEATE] Embedding cache: {cache_path}", file=sys.stderr)
+
+        # Initialize strategies
+        self.strategies = [
+            (NovelExplorationStrategy(config, self.csv), config.novel_exploration),
+            (HillClimbingStrategy(config, self.csv), config.hill_climbing),
+            (StructuralMutationStrategy(config, self.csv), config.structural_mutation),
+            (CrossoverStrategy(config, self.csv), config.crossover_hybrid),
+        ]
+
+    def get_context(self) -> IdeationContext:
+        """Build ideation context."""
+        with EvolutionCSV(self.config.csv_path) as csv:
+            top_performers = csv.get_top_performers(self.config.num_elites)
+            existing_descriptions = csv.get_all_descriptions()
+            generation = csv.get_highest_generation() + 1
+
+        # Read brief
+        brief_content = ""
+        if Path(self.config.brief_path).exists():
+            brief_content = Path(self.config.brief_path).read_text()[:1000]
+
+        return IdeationContext(
+            generation=generation,
+            top_performers=top_performers,
+            brief_content=brief_content,
+            existing_descriptions=existing_descriptions,
+            config=self.config
+        )
+
+    def check_novelty(self, description: str, existing: List[str]) -> Tuple[bool, float]:
+        """Check if description is novel enough."""
+        if not self.config.novelty_enabled:
+            return True, 0.0
+
+        if not existing:
+            return True, 0.0
+
+        try:
+            is_novel, max_sim = check_embedding_novelty(
+                description,
+                existing,
+                threshold=self.config.novelty_threshold
+            )
+            return is_novel, max_sim
+        except Exception as e:
+            print(f"[IDEATE] Novelty check failed: {e}", file=sys.stderr)
+            return True, 0.0  # Allow if check fails
+
+    def run(self) -> int:
+        """Run ideation. Returns number of ideas generated."""
+        context = self.get_context()
+        print(f"[IDEATE] Starting generation {context.generation}", file=sys.stderr)
+        print(f"[IDEATE] Top performers: {len(context.top_performers)}", file=sys.stderr)
+
+        all_ideas: List[Idea] = []
+        strategies_succeeded = 0
+
+        for strategy, count in self.strategies:
+            if count <= 0:
+                continue
+
+            ideas = strategy.generate(context, count)
+
+            if ideas:
+                strategies_succeeded += 1
+
+                # Filter for novelty
+                novel_ideas = []
+                for idea in ideas:
+                    is_novel, similarity = self.check_novelty(
+                        idea.description,
+                        context.existing_descriptions + [i.description for i in all_ideas]
+                    )
+
+                    if is_novel:
+                        novel_ideas.append(idea)
+                        print(f"[IDEATE] Accepted: {idea.id} (sim={similarity:.2%})", file=sys.stderr)
+                    else:
+                        print(f"[IDEATE] Rejected (too similar {similarity:.2%}): {idea.description[:50]}...", file=sys.stderr)
+
+                all_ideas.extend(novel_ideas)
+
+        # Add ideas to CSV
+        if all_ideas:
+            with EvolutionCSV(self.config.csv_path) as csv:
+                candidates = [
+                    {
+                        'id': idea.id,
+                        'basedOnId': idea.based_on_id,
+                        'description': idea.description,
+                        'status': 'pending',
+                        'idea-LLM': idea.strategy
+                    }
+                    for idea in all_ideas
+                ]
+                added = csv.append_candidates(candidates)
+                print(f"[IDEATE] Added {added} ideas to CSV", file=sys.stderr)
+
+        print(f"[IDEATE] Strategies succeeded: {strategies_succeeded}/{len([s for s, c in self.strategies if c > 0])}", file=sys.stderr)
+        print(f"[IDEATE] Total ideas generated: {len(all_ideas)}", file=sys.stderr)
+
+        # Final cache save
+        if self.config.novelty_enabled:
+            save_cache()
+
+        return len(all_ideas)
+
+
+def load_config(config_path: Optional[str] = None) -> IdeationConfig:
+    """Load configuration from YAML."""
+    import yaml
+
+    # Find config
+    if config_path:
+        yaml_path = Path(config_path)
+    elif os.environ.get('CLAUDE_EVOLVE_CONFIG'):
+        yaml_path = Path(os.environ['CLAUDE_EVOLVE_CONFIG'])
+    else:
+        yaml_path = Path('evolution/config.yaml')
+        if not yaml_path.exists():
+            yaml_path = Path('config.yaml')
+
+    if not yaml_path.exists():
+        raise FileNotFoundError(f"Config not found: {yaml_path}")
+
+    with open(yaml_path) as f:
+        data = yaml.safe_load(f) or {}
+
+    base_dir = yaml_path.parent
+
+    def resolve(path: str) -> str:
+        p = Path(path)
+        if not p.is_absolute():
+            p = base_dir / p
+        return str(p.resolve())
+
+    ideation = data.get('ideation', {})
+    novelty = data.get('novelty', {})
+
+    return IdeationConfig(
+        csv_path=resolve(data.get('csv_file', 'evolution.csv')),
+        evolution_dir=str(base_dir.resolve()),
+        brief_path=resolve(data.get('brief_file', 'BRIEF.md')),
+        algorithm_path=resolve(data.get('algorithm_file', 'algorithm.py')),
+        total_ideas=ideation.get('total_ideas', 15),
+        novel_exploration=ideation.get('novel_exploration', 3),
+        hill_climbing=ideation.get('hill_climbing', 5),
+        structural_mutation=ideation.get('structural_mutation', 3),
+        crossover_hybrid=ideation.get('crossover_hybrid', 4),
+        num_elites=ideation.get('num_elites', 3),
+        novelty_enabled=novelty.get('enabled', True),
+        novelty_threshold=novelty.get('threshold', 0.92)
+    )
+
+
+def main():
+    parser = argparse.ArgumentParser(description='Claude Evolve Ideation')
+    parser.add_argument('--config', help='Path to config.yaml')
+    parser.add_argument('--count', type=int, help='Override total idea count')
+    args = parser.parse_args()
+
+    try:
+        config = load_config(args.config)
+
+        ideator = Ideator(config)
+        count = ideator.run()
+
+        if count > 0:
+            print(f"Generated {count} ideas", file=sys.stderr)
+            sys.exit(0)
+        else:
+            print("No ideas generated", file=sys.stderr)
+            sys.exit(1)
+
+    except FileNotFoundError as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(1)
+    except Exception as e:
+        print(f"Error: {e}", file=sys.stderr)
+        import traceback
+        traceback.print_exc()
+        sys.exit(1)
+
+
+if __name__ == '__main__':
+    main()