boc/atm-anomaly-detection/scripts/create_instance.py

"""
VIMS Instance Creator

Creates a new VIMS instance for any article/topic.
Usage:
    python scripts/create_instance.py \
        --name "street-lighting" \
        --display-name "Street Lighting Monitoring" \
        --classes "pole_damage,light_out,vegetation_obstruction,vandalism" \
        --article-url "/insights/evidence-driven-municipal-maintenance/"
"""

import os
import argparse
from pathlib import Path


def create_instance(
    name: str,
    display_name: str,
    classes: str,
    article_url: str,
    base_dir: str = "/home/bernt/.openclaw/workspace/vims-core/instances"
):
    """
    Create new VIMS instance.
    
    Args:
        name: Instance name (directory name)
        display_name: Human-readable name
        classes: Comma-separated anomaly classes
        article_url: Related Landvex article URL
        base_dir: Base directory for instances
    """
    instance_dir = Path(base_dir) / name
    instance_dir.mkdir(parents=True, exist_ok=True)
    
    # Create subdirectories
    (instance_dir / "data" / "raw").mkdir(parents=True, exist_ok=True)
    (instance_dir / "data" / "processed").mkdir(parents=True, exist_ok=True)
    (instance_dir / "data" / "annotations").mkdir(parents=True, exist_ok=True)
    (instance_dir / "models").mkdir(exist_ok=True)
    (instance_dir / "src").mkdir(exist_ok=True)
    
    class_list = [c.strip() for c in classes.split(",")]
    
    # Create detector module
    detector_code = f'''"""
{name} Anomaly Detector
Generated by VIMS Instance Creator
Related article: {article_url}
"""

import sys
from pathlib import Path
sys.path.append(str(Path(__file__).parent.parent.parent / "core"))

from base_detector import VIMSBaseDetector, VIMSInstanceRegistry


class {name.title().replace("-", "")}Detector(VIMSBaseDetector):
    """
    Anomaly detector for {display_name}.
    
    Article: {article_url}
    """
    
    TOPIC = "{name}"
    CLASS_NAMES = {{
        {', '.join([f'{i}: "{c}"' for i, c in enumerate(class_list)])}
    }}
    
    SEVERITY_MAP = {{
        {', '.join([f'"{c}": 3' for c in class_list])}
    }}
    
    def preprocess(self, image):
        """{name}-specific preprocessing."""
        # TODO: Implement specific preprocessing
        return image
    
    def postprocess(self, raw_output):
        """{name}-specific postprocessing."""
        # TODO: Implement specific postprocessing
        return raw_output


# Register instance
VIMSInstanceRegistry.register("{name}", {name.title().replace("-", "")}Detector)
'''
    
    (instance_dir / "src" / "detector.py").write_text(detector_code)
    
    # Create database setup
    db_code = f'''"""
Database setup for {display_name}
"""

import sys
from pathlib import Path
sys.path.append(str(Path(__file__).parent.parent.parent / "core"))

from database import VIMSDatabase


def setup():
    """Initialize database for {name}."""
    db = VIMSDatabase("{name}")
    db.create_schema(anomaly_classes={class_list})
    print(f"Database initialized for {display_name}")


if __name__ == "__main__":
    setup()
'''
    
    (instance_dir / "src" / "database.py").write_text(db_code)
    
    # Create README
    readme = f'''# {display_name}

VIMS instance for {name}.

## Related Article
[{article_url}](https://landvex.com{article_url})

## Anomaly Classes
{chr(10).join([f"- {c}" for c in class_list])}

## Quick Start

1. Add training images to `data/raw/`
2. Annotate using LabelImg (YOLO format)
3. Run preprocessing: `python src/detector.py`
4. Train model: `python src/detector.py --train`
5. Run inference: `python src/detector.py --predict data/test/image.jpg`

## API

Once deployed, access via:
- REST: `POST /api/{name}/predict`
- WebSocket: `ws://host/ws/{name}/alerts`
'''
    
    (instance_dir / "README.md").write_text(readme)
    
    # Create config
    config = f'''# {name} configuration

topic: {name}
display_name: {display_name}
article_url: {article_url}

anomaly_classes:
{chr(10).join([f"  - {c}" for c in class_list])}

model:
  base: yolov8n.pt
  input_size: 640
  
training:
  epochs: 100
  batch_size: 16
'''
    
    (instance_dir / "config.yaml").write_text(config)
    
    print(f"✅ Created VIMS instance: {name}")
    print(f"   Location: {instance_dir}")
    print(f"   Classes: {', '.join(class_list)}")
    print(f"   Article: {article_url}")
    print()
    print("Next steps:")
    print(f"  1. cd {instance_dir}")
    print("  2. Add training images to data/raw/")
    print("  3. python src/database.py")
    print("  4. python src/detector.py --train")


def main():
    parser = argparse.ArgumentParser(description="Create VIMS Instance")
    parser.add_argument("--name", required=True, help="Instance name (directory)")
    parser.add_argument("--display-name", required=True, help="Human-readable name")
    parser.add_argument("--classes", required=True, help="Comma-separated anomaly classes")
    parser.add_argument("--article-url", required=True, help="Related article URL")
    
    args = parser.parse_args()
    
    create_instance(
        name=args.name,
        display_name=args.display_name,
        classes=args.classes,
        article_url=args.article_url
    )


if __name__ == "__main__":
    main()
'''

    (instance_dir / "src" / "detector.py").write_text(detector_code)
    
    # Create database setup
    db_code = f'''"""
Database setup for {display_name}
"""

import sys
from pathlib import Path
sys.path.append(str(Path(__file__).parent.parent.parent / "core"))

from database import VIMSDatabase


def setup():
    """Initialize database for {name}."""
    db = VIMSDatabase("{name}")
    db.create_schema(anomaly_classes={class_list})
    print(f"Database initialized for {display_name}")


if __name__ == "__main__":
    setup()
'''
    
    (instance_dir / "src" / "database.py").write_text(db_code)
    
    # Create README
    readme = f'''# {display_name}

VIMS instance for {name}.

## Related Article
[{article_url}](https://landvex.com{article_url})

## Anomaly Classes
{chr(10).join([f"- {c}" for c in class_list])}

## Quick Start

1. Add training images to `data/raw/`
2. Annotate using LabelImg (YOLO format)
3. Run preprocessing: `python src/detector.py`
4. Train model: `python src/detector.py --train`
5. Run inference: `python src/detector.py --predict data/test/image.jpg`

## API

Once deployed, access via:
- REST: `POST /api/{name}/predict`
- WebSocket: `ws://host/ws/{name}/alerts`
'''
    
    (instance_dir / "README.md").write_text(readme)
    
    # Create config
    config = f'''# {name} configuration

topic: {name}
display_name: {display_name}
article_url: {article_url}

anomaly_classes:
{chr(10).join([f"  - {c}" for c in class_list])}

model:
  base: yolov8n.pt
  input_size: 640
  
training:
  epochs: 100
  batch_size: 16
'''
    
    (instance_dir / "config.yaml").write_text(config)
    
    print(f"✅ Created VIMS instance: {name}")
    print(f"   Location: {instance_dir}")
    print(f"   Classes: {', '.join(class_list)}")
    print(f"   Article: {article_url}")
    print()
    print("Next steps:")
    print(f"  1. cd {instance_dir}")
    print("  2. Add training images to data/raw/")
    print("  3. python src/database.py")
    print("  4. python src/detector.py --train")


def main():
    parser = argparse.ArgumentParser(description="Create VIMS Instance")
    parser.add_argument("--name", required=True, help="Instance name (directory)")
    parser.add_argument("--display-name", required=True, help="Human-readable name")
    parser.add_argument("--classes", required=True, help="Comma-separated anomaly classes")
    parser.add_argument("--article-url", required=True, help="Related article URL")
    
    args = parser.parse_args()
    
    create_instance(
        name=args.name,
        display_name=args.display_name,
        classes=args.classes,
        article_url=args.article_url
    )


if __name__ == "__main__":
    main()