Interpret a Trained Tabular Model

Interpret a trained tabular model — a built-in Statistics and Machine Learning Toolbox (SMLT) classifier or regressor, or a custom model reached through a prediction function handle — by matching the user's intent to the right techniques, handling the custom-model case correctly, and avoiding the failure modes agents fall into on this workflow. This skill bundles the workflow in references/ (per-technique procedures and branch tables read on-demand) and scripts/ (the functionHandleMap helper). Do not invoke files in references/ as separate skills — they are only read when the corresponding step runs.

When to Use

• "Which predictors matter / feature importance / what drives the model"
• "Why did the model predict X for this case / explain this prediction"
• "What shape is the effect of predictor P / partial dependence / how does the prediction change with P"
• "Why did the model get these cases wrong / diagnose these errors / what would flip this"
• "Does the model behave differently for group Y / segment / cohort"
• Any of the above over a custom / non-SMLT model (a dlnetwork or a prediction function handle)

When NOT to Use