MorbidCorp/thermal-mortality-model

Thermal-Enhanced Morbid AI Model v0.1.1

Model Description

Initial THRML integration with basic probabilistic mortality modeling capabilities. Includes uncertainty quantification and demographic factor interactions.

This model integrates THRML (Thermodynamic HypergRaphical Model Library) with Morbid AI's mortality prediction capabilities, providing probabilistic predictions with uncertainty quantification.

Model Architecture

Type: thermal_energy_based_model Framework: THRML + JAX Version: 0.1.1

Thermal Features

• Probabilistic graphical models for mortality factors
• Block Gibbs sampling with demographic blocking
• Energy-based life expectancy prediction
• Confidence intervals for all predictions
• Risk factor analysis and contribution scoring

Performance Metrics

• baseline_accuracy: 0.8500
• uncertainty_coverage: 0.9500
• demographic_factors: 4.0000
• sampling_efficiency: 0.9200

Usage

from thermal.models.life_expectancy import LifeExpectancyEBM
from thermal.graph.mortality_graph import MortalityRecord

# Load mortality data
mortality_data = [...]  # List of MortalityRecord objects

# Initialize thermal model
model = LifeExpectancyEBM(mortality_data)

# Make prediction with uncertainty quantification
prediction = model.predict_life_expectancy(
    age=45,
    country="USA", 
    sex=1,  # 1=male, 2=female, 3=both
    n_samples=1000,
    confidence_level=0.95
)

print(f"Life Expectancy: {prediction.mean_life_expectancy:.1f} years")
print(f"95% CI: {prediction.confidence_interval}")
print(f"Uncertainty: {prediction.uncertainty:.2f}")

Model Configuration

THRML Parameters

sampling:

• default_samples: 1000
• burn_in: 200
• thinning: 2
• blocking_strategy: demographic model:
• energy_based: True
• uncertainty_quantification: True
• demographic_interactions: True performance:
• gpu_acceleration: True
• jax_backend: True
• memory_efficient: True

Sampling Configuration

• Block Gibbs Sampling: Two-color and demographic blocking strategies
• Default Samples: 1000 MCMC samples
• Burn-in: 200 steps
• Thinning: Every 2nd sample

Training Data

The model is trained on mortality data including:

• Countries: Global mortality statistics from major countries
• Age Range: 0-100+ years
• Time Period: 2010-2025
• Demographic Factors: Age, sex, country, year

Limitations

• Model performance depends on availability of demographic-specific training data
• Uncertainty estimates are calibrated on historical data and may not capture unprecedented events
• Requires THRML and JAX dependencies for optimal performance

Version History

v0.1.1 - 2025-10-29

• Initial THRML integration framework
• MortalityGraphBuilder for demographic interactions
• LifeExpectancyEBM with uncertainty quantification
• Block Gibbs sampling implementation
• Basic API integration structure

Citation

@software{thermal_morbid_ai_0_1_1,
  title={Thermal-Enhanced Morbid AI Model},
  version={0.1.1},
  year={2025},
  url={https://huggingface.co/MorbidCorp/thermal-mortality-model}
}

License

MIT License - see LICENSE file for details.

Contact

For questions about this model, please open an issue in the Morbid AI repository.