Metadata-Version: 2.1
Name: traauto
Version: 0.1.0
Summary: AutoML with Enhanced Track-Rail Algorithm, meta-learning, feature engineering, explainability, and smart ensemble fusion.
Home-page: https://github.com/eswaroy/automl_tra
Author: Dasari Ranga Eswar
Author-email: Dasari Ranga Eswar <rangaeswar890@gmail.com>
License: MIT
Classifier: Programming Language :: Python :: 3.7
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Operating System :: OS Independent
Classifier: License :: OSI Approved :: MIT License
Requires-Python: >=3.7
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy
Requires-Dist: pandas
Requires-Dist: scikit-learn
Requires-Dist: joblib
Requires-Dist: psutil
Provides-Extra: imblearn
Requires-Dist: imbalanced-learn; extra == "imblearn"
Provides-Extra: lightgbm
Requires-Dist: lightgbm; extra == "lightgbm"
Provides-Extra: lime
Requires-Dist: lime; extra == "lime"
Provides-Extra: optuna
Requires-Dist: optuna>=2.0.0; extra == "optuna"
Provides-Extra: shap
Requires-Dist: shap; extra == "shap"
Provides-Extra: torch
Requires-Dist: torch; extra == "torch"
Provides-Extra: xgboost
Requires-Dist: xgboost; extra == "xgboost"

# traauto: Advanced AutoML with Enhanced Track-Rail Algorithm

## Overview

**traauto** is a comprehensive, research-grade AutoML library that builds upon the success of the original [TRA Algorithm (`tra-algorithm`)](https://pypi.org/project/tra-algorithm/), extending the Track/Rail ensemble strategy with full automation, advanced feature engineering, meta-learning, dynamic model fusion, and explainable AI. This package is designed for practitioners and researchers who need both flexibility and state-of-the-art automation for classification or regression tasks, supporting modern datasets and robust analysis pipelines.

## About TRA Algorithm (`tra-algorithm`)

The original [TRA Algorithm](https://github.com/eswaroy/tra_algorithm) is an innovative ensemble framework that routes data between specialized models ("tracks") using real-time signals, enabling adaptive, explainable, parallel processing for both classification and regression. Key features included:

- Dynamic track switching & adaptive routing
- Advanced signal-based decision logic
- Track pruning and usage optimization
- Robust visualization and reporting APIs
- Full support for regression and classification tasks

For more details and usage of the original TRA Algorithm, see the [tra-algorithm documentation](https://pypi.org/project/tra-algorithm/).

## About traauto

**traauto** takes the core concept of dynamic track-based ensembles and delivers a full AutoML pipeline with these major enhancements:

- Automated Feature Engineering: Generation, selection, and transformation of features with adaptive regularization, interaction, transformation, and aggregation.
- Meta-Learning & Model Selection: Suggests architectures and hyperparameters using dataset meta-features and Optuna-based optimization.
- Smart Routing Engine: Intelligent per-record routing to specialized tracks, using meta-analytics, confidence, and feature specialization.
- Dynamic Ensemble Fusion: Advanced fusion strategies (stacking, weighted voting, smart fusion) optimize overall predictions dynamically.
- Regional Track Creation: Automated clustering to create regional/subtrack models for data specialization.
- Explainability: Supports SHAP, LIME, and feature importances for interpretability of predictions and model behaviour.
- Resource Monitoring and Adaptive Checkpoints: Tracks performance, memory usage, and provides checkpoint/resume capability.
- System Health & Analysis: Rich system summaries, track evaluations, routing analytics, and report generation.
- Project Persistence: Built-in checkpointing and reload via joblib for pipeline reproducibility.

## Installation

Install from PyPI:
```bash
pip install traauto
```
# Quick Start: Classification Example
```python

from traauto import EnhancedTrackRailSystem, TRAConfig
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
import numpy as np

def test_classification():
    # Generate data with compatible parameters
    X, y = make_classification(
        n_samples=120,
        n_features=10,
        n_classes=3,
        n_informative=3,
        random_state=0
    )
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)

    config = TRAConfig(task_type="classification")
    tra = EnhancedTrackRailSystem(config=config)
    tra.fit(X_train, y_train)
    y_pred = tra.predict(X_test)

    assert len(y_pred) == len(y_test)
    print("Classification test passed.")
    metrics = tra.get_performance_metrics(X_test, y_test)
    print("Classification accuracy:", metrics.get("overall_accuracy", "N/A"))
    print("F1-score:", metrics.get("f1_score", "N/A"))

if __name__ == "__main__":
    test_classification()
    print("-" * 40)

```

## Main Components and Capabilities

- **EnhancedTrackRailSystem:** Main system class combining all automation and analysis engines.
- **TRAConfig:** Configuration object for defining task type, feature engineering, meta-learning, resource constraints, and fusion strategies.
- **AutomatedFeatureEngineer:** Generates, transforms, and selects features adaptively.
- **MetaLearningEngine:** Analyzes dataset meta-features and applies hyperparameter optimization.
- **SmartRoutingEngine:** Dynamically chooses the best model/track per input using auditing.
- **DynamicEnsembleFusion:** Combines track outputs with weighted, voting, stacking, or smart fusion methods.
- **ExplainabilityEngine:** Produces SHAP/LIME/feature importance explanations for model or record-level predictions.
- **Checkpointing:** Safely saves and reloads full system state for reproducibility.

## Algorithm Architecture

1. Feature Engineering: The system automatically performs feature selection, interaction, transformation, and aggregation.
2. Track/Model Training: Multiple tracks (ensembles) are created for global and regional data distributions. Each receives optimized, specialized training.
3. Meta-Learning: Dataset statistics are extracted to suggest and tune model architectures and fusion strategies.
4. Routing & Prediction: For each new record, automl-tra routes data to the most suitable track/model using feature-based analytics and confidence.
5. Ensemble Fusion: Track predictions are combined using quality-weighted or stacking methods, adapting to dataset and model performance.
6. Explainability: Explanations are generated on-demand for individual predictions, tracks, or overall system behaviour.
7. Analysis & Reporting: System health, track statistics, and rich routing analysis are readily available.
8. Project Persistence: Full checkpoint and reload capability for workflows and reproducibility.

## Advanced Features

- Automated Clustering for Regional Tracks: High quality clustering (KMeans, DBSCAN) creates specialized regional tracks for improved accuracy.
- Adaptive Signal and Routing Creation: Meta-learned signals guide the routing engine to maximize predictive performance.
- Explainable AI Support: SHAP, LIME, and fallback importances provide interpretability for output and decisions.
- System Health Checks and Monitoring: Monitor latency, memory, stability, and prediction efficiency.
- Checkpoint/Restore: Robust project checkpointing for safe experiment pause/resume or deployment.
- Full Support for Imbalanced Data: Optional SMOTE integration and automatic class weight generation.

## Comparison: tra-algorithm vs automl-tra

| Feature                         | tra-algorithm                                    | traauto (this package)                  |
|----------------------------------|--------------------------------------------------|--------------------------------------------|
| Ensemble Track Routing           | Dynamic track switching                          | Advanced, meta-learned, per-record         |
| Signal Evaluation                | Adaptive, parallel signals                       | Meta-driven, smart routing engine          |
| AutoML Pipeline                  | Manual feature prep                              | Automated feature engineering, NAS         |
| Meta-Learning & Architecture     | Static models                                    | Automated model suggestion & optimization  |
| Feature Engineering              | Basic (scikit-learn)                             | Advanced, regularized & aggregated         |
| Dynamic Ensemble Fusion          | Voting/weighted fusion                           | Smart fusion, stacking, advanced fusion    |
| Explainable AI                   | Track-level                                      | Track-level & record-level (SHAP/LIME)     |
| Checkpointing & Recovery         |                                                  | Yes                                        |
| System Health/Analysis           | Basic logging/reporting                          | Full system and track analytics            |
| Task Support                     | Classification/Regression                        | Both, with more robust architecture        |

## Requirements

- Python >= 3.7
- numpy, pandas, scikit-learn, joblib, psutil
- Optional (for full feature set): optuna, torch, xgboost, lightgbm, shap, lime, imbalanced-learn


## Support & Issues

- Email: rangaeswar890@gmail.com
- Issues/Discussions: [GitHub Issues](https://github.com/eswaroy/automl_tra/issues)

## License

This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.

## Citation

If you use automl-tra in your research, please cite:

