Orange3-MLflow-Export 0.1.2

Export Orange3 models with preprocessing pipelines to MLflow format for production deployment.

Orange3 MLflow Export

An Orange3 add-on that provides seamless export of complete machine learning workflows to MLflow format, including preprocessing pipelines and dependency management for reproducible deployment.

Overview

Export your Orange3 models with all preprocessing steps intact! This add-on bridges the gap between Orange3's visual machine learning environment and MLflow's production deployment capabilities.

Key Features

Complete Pipeline Export

• Preprocessing Included: Unlike basic model saving, captures entire preprocessing pipelines (normalization, imputation, feature selection, etc.)
• Domain Transformations: Preserves Orange's automatic domain handling for seamless prediction
• Zero Information Loss: Everything needed for prediction is included

Production-Ready Deployment

• MLflow Standard Format: Exports to industry-standard MLflow model format
• Dependency Management: Automatically includes Orange3, scikit-learn, pandas, numpy, and all dependencies
• Cross-Platform: Deploy on any platform supporting MLflow (local, cloud, containers)
• Serving Ready: Use mlflow models serve immediately after export

Smart Input Handling

• Flexible DataFrames: Accepts pandas DataFrames with automatic type conversion
• Column Name Mapping: Intelligently handles feature name mismatches through positional mapping
• Signature Inference: Automatically generates input/output schemas from sample data

Universal Compatibility

• All Orange Models: Works with any Orange learner (classification, regression, clustering)
• Any Preprocessing: Supports all Orange preprocessing widgets and operations
• Version Independent: Models work across different Orange and Python versions

Quick Start

Installation

pip install orange3-mlflow-export
# or from source
pip install .

Basic Usage

In Orange GUI:

1. Build your workflow (File → Preprocess → Model)
2. Add MLflow Export widget (Example section)
3. Connect your trained model
4. Set export path and save

Programmatically:

from orangecontrib.example.widgets.owmlflowexport import OWMLFlowExport
import Orange

# Train model with preprocessing
data = Orange.data.Table("iris")
preprocessor = Orange.preprocess.preprocess.PreprocessorList([
    Orange.preprocess.Normalize(),
    Orange.preprocess.Impute()
])
processed_data = preprocessor(data)
model = Orange.classification.LogisticRegressionLearner()(processed_data)

# Export to MLflow
widget = OWMLFlowExport()
widget.set_model(model)
widget.filename = "model.mlflow"
widget.do_save()

Deploy and Serve:

mlflow models serve -m ./model.mlflow -p 8080

Why Choose This Over Standard Orange Model Saving?

Feature	Orange Save Model	MLflow Export
Preprocessing	❌ Lost	✅ Preserved
Dependencies	❌ Manual	✅ Automatic
Deployment	❌ Complex	✅ One Command
Portability	❌ Limited	✅ Universal
Production Use	❌ Difficult	✅ Seamless

Use Cases

• Model Deployment: Export Orange models for production serving
• Reproducible Research: Share complete workflows with dependencies
• Cross-Platform Models: Train in Orange, deploy anywhere
• MLOps Integration: Integrate Orange models into MLflow tracking and serving
• Educational: Demonstrate complete ML pipeline deployment

Supported Orange Components

Models: All Orange learners (Tree, SVM, Neural Network, Linear, etc.)
Preprocessing: Normalize, Impute, Continuize, Feature Selection, PCA, etc.
Data Types: Tabular, spectroscopic, time series, text (with appropriate preprocessing)

Testing

Verify your exported models with included test scripts:

python test_spectra_model.py ./your_model.mlflow
python load_predictor.py

Requirements

• Python 3.8+
• Orange3
• MLflow
• pandas, numpy, scikit-learn
• cloudpickle

Documentation

See the complete documentation for detailed usage, architecture, and troubleshooting.

License

GPL-3.0 (same as Orange3)