Skip to navigation

Symbolic explanations

Generate and review a written interpretation of a module’s rules.

View as Markdown

Symbolic Explanations generates a written interpretation of a selected module’s decision rules. Use it to inspect learned module logic; use Local explanations for a particular observation.

Before you begin

Choose a running deployment and a module with decision rules to inspect. No individual observation is required. Check the rule conditions and expressions before asking for a narrative interpretation.

Generate a symbolic explanation

  1. Open Explanation Lab > Symbolic Explanations.

  2. Select the intended Deployed Model.

  3. Choose a Module. The example selects age.

  4. Review the displayed Decision Rules, including their condition, expression, and intercept.

  5. Select Generate Symbolic Explanation.

  6. Read the completed explanation alongside the source rule.

Interpret and review the result

The selected age module has one rule with condition All Values. Its displayed expression is 0.0*age² + (0.0039*age - 0.1580), and its intercept is approximately -0.1580. These are rounded values for the example model; use the values from your own selected module when interpreting its rule.

The example explanation includes an overview, model-structure summary, per-regime interpretation, cross-regime dynamics and rule references. Use Copy to copy the generated text, or Export to choose Markdown or PDF when the explanation is ready to reuse.

Symbolic Explanations interprets a selected module’s rules. It does not require an input value for a particular observation. Use Local Explanations when you need to explain a prediction for a specific record.

Review before reuse: verify that the narrative agrees with the rule, distinguishes a module contribution from the final prediction, and preserves the scale and rounding of any quoted values. A statement about increasing a module’s contribution should not automatically become a claim about a causal effect or the complete model’s output.

The age module rule used to generate a symbolic explanation.

The age module rule used to generate a symbolic explanation.

Generated symbolic text. Verify statements about predictions against the selected module rule.

Generated symbolic text. Verify statements about predictions against the selected module rule.

Copy or export a reviewed explanation

After checking the narrative against the selected rule, use Copy or Export. Retain the deployment, module name and source rule with the generated text. When a coefficient is rounded on screen, preserve that rounding in the narrative instead of presenting extra precision.

Next steps

Continue to Use explanations.