demo.py

"""
Demo script for Bio-ACDC framework.
Runs a minimal coevolution experiment with small models on CPU.
"""

import logging
import os
import sys

# Setup logging
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
    handlers=[logging.StreamHandler(sys.stdout)]
)

from bio_acdc import BioACDC, BioACDCConfig
from bio_acdc.tasks import BioTaskPool
from bio_acdc.mergers import LinearMerge
from bio_acdc.mutators import GaussianNoiseMutator
from bio_acdc.evaluator import BioEvaluator
from bio_acdc.archive import BioArchive

def demo_task_pool():
    """Demonstrate task pool functionality."""
    print("\n" + "="*60)
    print("DEMO 1: Task Pool")
    print("="*60)
    
    pool = BioTaskPool(seed=42)
    print(f"\nInitial pool has {len(pool.tasks)} tasks")
    
    # Sample tasks
    tasks = pool.get_tasks(n=5)
    for i, task in enumerate(tasks):
        print(f"\nTask {i+1}: {task.task_id}")
        print(f"  Type: {task.task_type}, Family: {task.task_family}")
        print(f"  Prompt: {task.prompt[:80]}...")
        print(f"  Difficulty: {task.difficulty:.2f}")
        print(f"  Expected: {task.expected_answer}")
    
    # Generate new tasks
    archive = BioArchive(archive_size=5)
    new_tasks = pool.generate_new_tasks(archive, num_tasks=3)
    print(f"\nGenerated {len(new_tasks)} new tasks")
    print(f"Pool size now: {len(pool.tasks)}")
    
    return pool


def demo_archive():
    """Demonstrate archive functionality."""
    print("\n" + "="*60)
    print("DEMO 2: Dominated Novelty Search Archive")
    print("="*60)
    
    archive = BioArchive(archive_size=5, k_neighbors=2)
    
    # Simulate adding solutions
    from bio_acdc.archive import BioSolution
    
    solutions = [
        BioSolution("model_A", 0.7, {"task_1": 0.8, "task_2": 0.6, "task_3": 0.9}, generation=1),
        BioSolution("model_B", 0.6, {"task_1": 0.5, "task_2": 0.7, "task_3": 0.5}, generation=1),
        BioSolution("model_C", 0.5, {"task_1": 0.9, "task_2": 0.4, "task_3": 0.3}, generation=1),
    ]
    
    for sol in solutions:
        archive.add_solution(sol)
    
    print(f"\nAdded {len(solutions)} solutions to archive")
    print(f"Coverage: {archive.get_coverage():.2%}")
    
    best = archive.get_best()
    print(f"Best solution: {best.model_path} (fitness={best.fitness:.3f})")
    
    # Add new solutions
    new_solutions = [
        BioSolution("model_D", 0.65, {"task_1": 0.6, "task_2": 0.8, "task_3": 0.4}, generation=2),
        BioSolution("model_E", 0.55, {"task_1": 0.3, "task_2": 0.9, "task_3": 0.2}, generation=2),
    ]
    
    archive.update(new_solutions)
    print(f"\nAfter update: {len(archive.solutions)} solutions")
    
    best = archive.get_best()
    print(f"New best: {best.model_path} (fitness={best.fitness:.3f})")
    print(f"Coverage: {archive.get_coverage():.2%}")


def demo_merging():
    """Demonstrate model merging with toy models."""
    print("\n" + "="*60)
    print("DEMO 3: Model Merging (Toy Example)")
    print("="*60)
    
    import torch
    import json
    from safetensors.torch import save_file
    
    # Create two toy models
    model_dir = "/tmp/toy_models"
    os.makedirs(model_dir, exist_ok=True)
    
    # Create simple state dicts
    model_a_weights = {
        "layer1.weight": torch.randn(10, 10) * 0.1,
        "layer1.bias": torch.randn(10) * 0.1,
    }
    model_b_weights = {
        "layer1.weight": torch.randn(10, 10) * 0.1,
        "layer1.bias": torch.randn(10) * 0.1,
    }
    
    model_a_path = os.path.join(model_dir, "model_a")
    model_b_path = os.path.join(model_dir, "model_b")
    
    os.makedirs(model_a_path, exist_ok=True)
    os.makedirs(model_b_path, exist_ok=True)
    
    save_file(model_a_weights, os.path.join(model_a_path, "model.safetensors"))
    save_file(model_b_weights, os.path.join(model_b_path, "model.safetensors"))
    
    # Write config
    config = {"num_layers": 1, "hidden_size": 10}
    with open(os.path.join(model_a_path, "config.json"), "w") as f:
        json.dump(config, f)
    with open(os.path.join(model_b_path, "config.json"), "w") as f:
        json.dump(config, f)
    
    print(f"\nCreated toy models at {model_dir}")
    
    # Linear merge
    from bio_acdc.mergers import LinearMerge
    merger = LinearMerge(weights=[0.6, 0.4])
    merged_path = merger.merge(
        parent_paths=[model_a_path, model_b_path],
        save_path=os.path.join(model_dir, "merged"),
    )
    
    print(f"Merged model saved to: {merged_path}")
    
    # Verify merge
    from safetensors.torch import load_file
    merged = load_file(os.path.join(merged_path, "model.safetensors"))
    print(f"Merged weights shape: {merged['layer1.weight'].shape}")
    
    # Mutation
    from bio_acdc.mutators import GaussianNoiseMutator
    mutator = GaussianNoiseMutator(std=0.01)
    mutated_path = mutator.mutate(merged_path, os.path.join(model_dir, "mutated"))
    
    print(f"Mutated model saved to: {mutated_path}")


def demo_bio_acdc_mock():
    """Demonstrate full Bio-ACDC loop with mock components."""
    print("\n" + "="*60)
    print("DEMO 4: Full Bio-ACDC Loop (Mock)")
    print("="*60)
    
    from bio_acdc.core import BioACDCConfig
    from bio_acdc.archive import BioSolution, BioArchive
    
    # Mock components that don't need real models
    class MockTaskPool:
        def __init__(self):
            self.tasks = []
            self.rng = __import__('numpy').random.RandomState(42)
        def get_tasks(self, n=10):
            return []
        def generate_new_tasks(self, archive, num_tasks=5, difficulty_weights=None):
            return []
    
    class MockEvaluator:
        def evaluate_model(self, model_path, tasks):
            # Return random scores
            import random
            return {f"task_{i}": random.random() for i in range(10)}
    
    class MockMerger:
        def merge(self, parent_paths, save_path, base_model_path=None):
            os.makedirs(save_path, exist_ok=True)
            return save_path
    
    class MockMutator:
        def mutate(self, model_path, save_path):
            return save_path
    
    config = BioACDCConfig(
        archive_size=10,
        num_generations=3,
        offspring_per_gen=2,
        seed_model_paths=["/tmp/model_a", "/tmp/model_b"],
        output_dir="/tmp/bio_acdc_mock",
    )
    
    task_pool = MockTaskPool()
    evaluator = MockEvaluator()
    merger = MockMerger()
    mutator = MockMutator()
    
    print("\nRunning mock Bio-ACDC for 3 generations...")
    
    # Simulate
    archive = BioArchive(archive_size=10)
    
    for gen in range(1, 4):
        print(f"\n--- Generation {gen} ---")
        
        # Mock offspring
        new_sols = []
        for i in range(2):
            sol = BioSolution(
                model_path=f"/tmp/gen_{gen}_ind_{i}",
                fitness=0.5 + 0.1 * (gen + i),
                skill_vector={f"task_{j}": 0.5 + 0.1 * j for j in range(10)},
                generation=gen,
            )
            new_sols.append(sol)
            print(f"  Created offspring {i}: fitness={sol.fitness:.3f}")
        
        # Update archive
        if gen == 1:
            for sol in new_sols:
                archive.add_solution(sol)
        else:
            archive.update(new_sols)
        
        best = archive.get_best()
        print(f"  Archive size: {len(archive.solutions)}")
        if best:
            print(f"  Best: {best.model_path} (fitness={best.fitness:.3f})")
    
    print("\nMock evolution complete!")


def main():
    """Run all demos."""
    print("\n" + "="*60)
    print("BIO-ACDC FRAMEWORK DEMO")
    print("Biological Sequence Model Coevolution")
    print("="*60)
    
    demo_task_pool()
    demo_archive()
    demo_merging()
    demo_bio_acdc_mock()
    
    print("\n" + "="*60)
    print("ALL DEMOS COMPLETE!")
    print("="*60)
    print("\nBio-ACDC is ready for use with real biological LMs.")
    print("\nNext steps:")
    print("1. Install requirements: pip install -r requirements.txt")
    print("2. Download seed models (e.g., facebook/esm2_t33_650M_UR50D)")
    print("3. Run full coevolution: python -c 'from demo_bio_acdc import run_full; run_full()'")
    print("\nSee README.md for detailed usage instructions.")


if __name__ == "__main__":
    main()