Simulation Decomposition (SimDec) in Python: Visual Uncertainty & Sensitivity Analysis

Understanding uncertainty is no longer optional. Whether you are building AI systems, running engineering simulations, or evaluating real-world experiments, uncertainty and sensitivity determine how your model behaves—and how reliable your conclusions really are.

Simulation Decomposition (SimDec) is a new, visual, and accessible way to make sense of uncertainty. Recently published in the Journal of Open Source Software, SimDec introduces a Python API and a no-code dashboard that bring powerful uncertainty–sensitivity analysis tools to everyone.
Source: SimDec JOSS paper

Roy, P. T., & Kozlova, M. (2024). Simulation decomposition in Python. Journal of Open Source Software, 9(98), 6713.

What Is SimDec?

Traditional sensitivity analysis relies heavily on numerical indices such as Sobol’ indices to rank input importance. These methods are powerful, but they don’t show how inputs interact or what behavioral patterns emerge in the output.

SimDec goes beyond this by combining:

• Variance-based sensitivity indices
• Automatic scenario formation
• A visual decomposition of the output distribution

According to the paper, the method “reveals the critical behavior of a computational model or an empirical dataset” by automatically forming meaningful scenarios from the most influential inputs.
Source: page 1–2, SimDec JOSS paper

How scenarios are created

SimDec identifies the most important inputs using sensitivity indices.
It then:

1. Splits their numeric ranges into states (e.g., low, medium, high)
2. Creates an exhaustive list of combinations
3. Decomposes the output distribution accordingly

The result is a plot like the one on page 2 of the publication—a layered histogram that visually explains how different input states affect the output.
Source: Figure 1, SimDec JOSS paper

Why This Matters for AI, Engineering, and Policy

With the rise of AI regulation (e.g., EU AI Act) and the need for transparent, explainable models, uncertainty analysis is now mission-critical. SimDec addresses this need by offering:

• A human-readable explanation of uncertainty
• A transparent sensitivity and interaction view
• Tools that work for both computational models and empirical datasets
  Source: SimDec JOSS paper, Statement of Need section

This makes it especially valuable in:

• AI/ML model assessment
• Environmental and engineering simulations
• Cost-benefit analysis and policy evaluation
• Risk-sensitive decision-making

SimDec in Python: Accessible to Everyone

The published package offers:

• A clean, well-structured Python API
• Integration with SALib and SciPy for sensitivity analysis and Quasi-Monte Carlo (QMC) sampling
• A fully interactive web dashboard at https://simdec.io
  Source: page 2 of SimDec JOSS paper

This means users can start with no coding experience—or automate workflows with a few lines of Python.

What the Visualization Tells You

The figure on page 2 shows an example:

• The most influential input splits the output into three clearly distinct states.
• A second input further subdivides them—but only matters when the first input is low.
• When the first input is medium or high, the second input has almost no effect.
  Source: Figure 1, SimDec JOSS paper

This kind of insight is extremely hard to obtain using numerical indices alone.

Conclusion

SimDec brings a new level of clarity to uncertainty and sensitivity analysis. It combines rigorous statistical foundations with intuitive visualization, making it easier than ever to understand how your model behaves—and why.

With both a Python package and a no-code dashboard, SimDec is built for researchers, practitioners, and decision-makers who want deeper insight into model behavior.

Explore the dashboard: https://simdec.io

‍