mlflow2rdf 0.1.2

YAML-config-driven MLflow tracking data to RDF knowledge graphs with MLSO ontology alignment

MLflow to RDF Converter

YAML-config-driven MLflow tracking data to RDF knowledge graphs — aligned with the MLSO ontology.

---

Overview

mlflow2rdf converts MLflow experiment/run/parameter/metric data into RDF triples using declarative YAML mappings. It supports two modes:

• RML Standard (recommended): W3C RML-compliant, aligned with MLSea
• Custom YAML: Simplified configuration for quick use

---

Project Structure

mlflow-to-rdf/
├── config/
│   ├── rml_mappings.yaml   # RML standard mappings (recommended) ⭐
│   ├── mappings.yaml        # Custom declarative mappings
│   ├── sources.yaml         # MLflow data source config
│   └── validation.yaml      # SHACL validation rules
├── src/
│   ├── converter_rml.py     # RML standard converter ⭐
│   ├── rml_engine.py       # RML engine wrapper
│   ├── data_collector.py   # MLflow data collector
│   ├── converter.py        # Main converter (CLI entry point)
│   ├── engine.py           # Declarative conversion engine
│   ├── validators.py       # SHACL validator
│   └── utils.py            # Utilities
├── docs/
│   ├── PROJECT_DOCUMENTATION.md
│   ├── COMPARISON_ANALYSIS.md
│   └── RML_IMPLEMENTATION_SUMMARY.md
├── tests/
├── examples/
└── data/

---

Quick Start

Installation

pip install mlflow2rdf

Basic Usage (CLI)

# Point to your MLflow tracking server
mlflow2rdf --mlflow-uri http://localhost:5000 --output output.ttl

Programmatic Usage

from mlflow2rdf import DeclarativeConverter, SHACLValidator

# Initialize converter with YAML configs
converter = DeclarativeConverter(
    sources_config_path='config/sources.yaml',
    mappings_config_path='config/mappings.yaml'
)

# Execute transformation
converter.convert()

# Validate against SHACL shapes
validator = SHACLValidator('config/validation.yaml')
result = validator.validate(converter.graph)

# Save RDF output
converter.save('output.ttl')

---

Configuration

config/sources.yaml — Data Source

mlflow:
  type: mlflow
  uri: http://localhost:5000   # MLflow tracking server URI
  api_version: 2.0
  extraction:
    experiments: { enabled: true }
    runs: { enabled: true, experiment_ids: ["0"] }
    params: { enabled: true }
    metrics: { enabled: true }
    tags: { enabled: true }

output:
  format: turtle
  path: ./data/output.ttl
  namespaces:
    mlso:   http://example.org/mlso/
    prov:   http://www.w3.org/ns/prov#
    dcterms: http://purl.org/dc/terms/
    rdfs:   http://www.w3.org/2000/01/rdf-schema#

config/mappings.yaml — Declarative Mapping Rules

Maps MLflow entities to MLSO ontology classes:

MLflow Entity	MLSO Class
Experiment	mlso:Experiment
Run	mlso:Run
Parameter	mlso:HyperParameterSetting
Metric	mlso:Metric
Tag	dcterms:hasPart

config/validation.yaml — SHACL Shapes

Defines constraints (minCount, maxCount, datatype, allowed values) for each MLSO class. Run custom SPARQL rules for cross-entity validation.

---

Features

• ✅ Declarative mappings — all transformation rules in YAML, no code changes needed
• ✅ RML standard support — W3C RML-compliant with YARRRML input
• ✅ MLSO ontology alignment — typed to MLSO vocabulary
• ✅ SHACL validation — ensure RDF output conforms to ontology constraints
• ✅ Multiple RDF serializations — Turtle, N3, JSON-LD, XML
• ✅ CLI and library — use as a tool or import as a package

---

Documentation

• README_RML.md — RML standard usage guide
• docs/PROJECT_DOCUMENTATION.md — Full technical documentation
• docs/COMPARISON_ANALYSIS.md — Analysis vs. MLSea
• docs/RML_IMPLEMENTATION_SUMMARY.md — RML implementation summary

---

Related Resources

• MLSea Paper: "MLSea: A Semantic Layer for Discoverable Machine Learning"
• MLSea GitHub: https://github.com/dtai-kg/MLSO
• RML Spec: https://rml.io/specs/rml/
• YARRRML Spec: https://rml.io/yarrrml/spec/

---

Author: Jason Jia
Version: 0.1.1