Explaining Model Predictions by Hand: Contributions, Shapley Values, Permutation Importance, Partial Dependence & Counterfactuals
Open up a model's predictions with nothing but a spreadsheet. You will break a linear score into feature contributions, compute exact Shapley values for a three-feature model from all eight coalitions, measure permutation importance, draw partial-dependence and individual curves, and find the smallest change that flips a decision, with every cell checked as you go.
One-time payment · lifetime access · lessons, labs, exam and certificate included · no subscription.
About this course
This course opens up toy models with nothing but a spreadsheet. You split a linear rent prediction into one contribution per feature and check that baseline plus contributions rebuilds it. For a three-feature risk score you build all eight coalitions, average marginal contributions over six orders into exact Shapley values and confirm they add up. You see how Shapley splits an interaction effect, measure permutation importance with shuffled columns and mean absolute error, draw ICE and partial-dependence curves, find the smallest change that flips a decision with MINIFS, and finish with a global importance chart, a stability check and the limits of every explanation.
Lessons
- Baseline & Contributions: Explaining a Linear Score Feature by Feature Pro
- Coalitions: All Eight Feature Subsets & the Model Output for Each Pro
- Shapley Values: Marginal Contributions over Every Order & the Weighted Formula Pro
- Interactions: How Shapley Splits an Effect Two Features Create Together Pro
- Permutation Importance: How Much Worse the Model Gets When a Column Is Shuffled Pro
- Partial Dependence & ICE Curves: One Feature Across the Whole Fleet Pro
- Counterfactuals: The Smallest Change That Flips a Decision Pro
- The Explanation Report: Global vs Local, Stability & Limitations Pro