Wrench 🔧

A powerful framework for building automated sensor registration pipelines.

Overview

Wrench is a modular, extensible workflow framework designed to streamline the process of harvesting, enriching, and registering sensor metadata from diverse IoT sources into urban data catalogs. It provides a standardized pipeline architecture with interchangeable components to help make sensor data more discoverable and valuable.

Features

• 🔄 Automated Metadata Harvesting: Extract metadata from various IoT data sources with minimal configuration
• 📊 Standardized Data Models: Type-safe data structures using Pydantic for consistent handling of metadata
• 🔍 Advanced Classification: Group similar sensors using machine learning and taxonomy-based approaches
• ✨ Metadata Enrichment: Enhance sensor descriptions with contextual information using LLM technologies
• 🏗️ Modular Architecture: Compose workflows from interchangeable components for maximum flexibility
• 🔌 Extensible Interfaces: Easily add support for new data sources and catalog systems
• 🤖 LLM Integration: Leverage AI capabilities for automatic content generation and classification

Installation

pip install auto-wrench

To install with specific component dependencies:

# For SensorThings support
pip install 'auto-wrench[sensorthings]'

# For KINETIC grouper
pip install 'auto-wrench[kinetic]'

# Multiple components
pip install 'auto-wrench[sensorthings,kinetic]'

Core Components

Wrench consists of four main component types that can be combined in a pipeline:

1. Harvesters: Extract metadata from IoT data sources (e.g., SensorThings API)
2. Groupers: Classify and organize sensors into meaningful groups using various ML approaches
3. MetadataEnrichers: Build spatial and temporal metadata for services and sensor groups
4. Catalogers: Register the processed metadata into data catalogs (e.g., SDDI/CKAN)

Each component type follows a standardized interface, making it easy to extend with custom implementations.

Quick start

The following example sets up a complete pipeline with a SensorThings API harvester, a KINETIC grouper for classification, and an SDDI cataloger for registration:

import asyncio

from wrench.cataloger.sddi import SDDICataloger
from wrench.grouper.kinetic import KINETIC
from wrench.harvester.sensorthings import SensorThingsHarvester
from wrench.metadataenricher.sensorthings import SensorThingsMetadataEnricher
from wrench.pipeline.sensor_pipeline import SensorRegistrationPipeline
from wrench.utils.config import LLMConfig

llm_config = LLMConfig(base_url="https://my-llm.com", model="llama3.3:70b-instruct-q4_K_M")

# Initialize components
harvester = SensorThingsHarvester(
    base_url="https://example.org/v1.1",
    pagination_config={"page_delay": 0.2, "timeout": 60, "batch_size": 100},
)
grouper = KINETIC(
    llm_config=llm_config,
    embedder="intfloat/multilingual-e5-large-instruct",
    resolution=1,
)
metadata_enricher = SensorThingsMetadataEnricher(
    base_url="https://example.org/v1.1",
    title="City Sensor Network",
    description="Environmental sensors across the city",
    llm_config=llm_config,
)
cataloger = SDDICataloger(
    base_url="https://catalog.example.org",
    api_key="your-api-key",
    owner_org="your-organization",
)

# Assemble and run the pipeline
pipeline = SensorRegistrationPipeline(
    harvester=harvester,
    grouper=grouper,
    metadataenricher=metadata_enricher,
    cataloger=cataloger,
)

result = asyncio.run(pipeline.run_async())

Configuration

Wrench supports YAML-based pipeline configuration through PipelineRunner.from_config_file(). The top-level template_ key selects the pipeline template. Environment variables are resolved using ${VAR_NAME} syntax.

# pipeline_config.yaml
template_: SensorPipeline

harvester:
  sensorthings:
    base_url: "https://example.org/v1.1"
    pagination_config:
      page_delay: 0.2
      timeout: 60
      batch_size: 100

grouper:
  kinetic:
    llm_config:
      model: ${OLLAMA_MODEL}
      base_url: ${OLLAMA_URL}
      api_key: ${OLLAMA_API_KEY}

metadataenricher:
  sensorthings:
    base_url: "https://example.org/v1.1"
    title: "City Sensor Network"
    description: "Environmental sensors across the city"
    llm_config:
      model: ${OLLAMA_MODEL}
      base_url: ${OLLAMA_URL}
      api_key: ${OLLAMA_API_KEY}

cataloger:
  noop: {}

Run a YAML-configured pipeline with:

import asyncio
from wrench.pipeline.config import PipelineRunner

runner = PipelineRunner.from_config_file("pipeline_config.yaml")
result = asyncio.run(runner.run({}))

Component Overview

Harvesters

Harvesters connect to data sources and extract metadata. Wrench includes:

• SensorThingsHarvester: Connects to OGC SensorThings API endpoints
• Extensible base class for creating custom harvesters

Groupers

Groupers organize sensors into logical groups using various machine learning approaches:

• KINETIC: Keyword-Informed, Network-Enhanced Topical Intelligence Classifier with hierarchical clustering
• LDAGrouper: Latent Dirichlet Allocation for topic modeling and device grouping
• BERTopicGrouper: BERTopic-based clustering with HDBSCAN and UMAP for topic discovery
• Can be extended with custom grouping algorithms

MetadataEnrichers

MetadataEnrichers build spatial and temporal metadata for items and groups:

• SensorThingsMetadataEnricher: Builds metadata for SensorThings API data sources
• Extensible base class for different data source types

Catalogers

Catalogers register metadata into data catalogs:

• SDDICataloger: Registers metadata into SDDI/CKAN-based catalogs
• Extensible interface for supporting other catalog systems

Advanced Features

Advanced grouping with ML

Different groupers offer various approaches for sensor classification:

from wrench.utils.config import LLMConfig

# KINETIC for hierarchical topic clustering
from wrench.grouper.kinetic import KINETIC
grouper = KINETIC(
    llm_config=LLMConfig(base_url="https://my-llm.com", model="llama3.3:70b-instruct-q4_K_M"),
    embedder="intfloat/multilingual-e5-large-instruct",
    lang="en",
    resolution=1,
)

# LDA for topic modeling (no extra dependencies required)
from wrench.grouper.lda import LDAGrouper
from wrench.grouper.lda.models import LDAConfig
grouper = LDAGrouper(config=LDAConfig(n_topics=10, alpha=0.1, beta=0.01))

# BERTopic for density-based clustering (requires sentence-transformers, hdbscan, umap-learn, bertopic)
from wrench.grouper.bertopic import BERTopicGrouper
from wrench.grouper.bertopic.models import BERTopicConfig
grouper = BERTopicGrouper(config=BERTopicConfig(min_topic_size=10))

Development

Setting up the Development Environment

# Clone the repository
git clone https://github.com/urbansense/wrench.git
cd wrench

# Run the make target for full setup
make setup

# Install component dependencies for development
uv pip install -e ".[sensorthings,kinetic]"

Code Style and Testing

This project follows the Ruff code style and uses comprehensive testing:

# Format and lint code
make format
make lint

# Run tests with coverage
make test

# Run specific test types
make test_unit
make test_e2e

# Type checking
make lint_types

Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Please ensure your code follows our coding standards and includes appropriate tests.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Support and Documentation

For support, please:

• Open an issue in the GitHub repository
• Check the documentation
• Contact the development team at jeffrey.limnardy@tum.de