AI Glossary
The complete dictionary of Artificial Intelligence
Individual Conditional Expectation (ICE) Plot
Graphical representation that plots the prediction for each individual observation as a function of a feature, revealing the heterogeneity of effects that PDP masks.
Shapley Plot
Visualization of each feature's contribution to a specific prediction, based on Shapley values from game theory to ensure fair allocation of impact.
SHAP Global Importance Plot
Bar chart that ranks features by their average impact on the magnitude of the model's prediction, calculated by taking the mean absolute value of SHAP values.
SHAP Beeswarm Plot
Visualization combining feature importance and their influence, where each point is a SHAP value for a prediction, positioned on the feature axis and colored by the feature value.
SHAP Interaction Plot
Visualization that highlights interaction effects between features, showing how the effect of one feature on the prediction depends on the value of another.
LIME Heatmap
Visual overlay on input data (such as an image or text) that highlights the segments or words that contributed most to a model's local prediction, according to the LIME method.
Surrogate Decision Tree
Visualization of a simple decision tree trained to mimic the behavior of a complex model (black box), providing an approximate global interpretation of the model's decision rules.
ALE Contour Plot
Two-dimensional visualization of the Accumulated Local Effect (ALE) for a pair of features, using contours to show how the prediction changes over the space of these two features.
Permutation Importance Diagram
Bar chart that measures and displays the decrease in a model's performance when the values of a single feature are randomly permuted, indicating its predictive importance.
Interpretability Learning Curve
Graph that plots the performance of an interpretability model (such as a surrogate) as a function of the amount of data or allowed complexity, helping to evaluate the fidelity-interpretability trade-off.
Counterfactual Visualization
Graphical representation that compares an original instance to its modified version (counterfactual) that changes the model's prediction, highlighting the features that were altered.
Surrogate Fidelity Graph
Scatter plot or curve that compares the predictions of the surrogate model (interpretable) to the predictions of the original model (black box) to assess the quality of the approximation.
Anchors Rules Diagram
Visualization that presents a set of conditions (the anchor) that sufficiently guarantee a prediction, often in the form of a hierarchical list or flow diagram for a local explanation.
Gradient Saliency Map
Heat map that overlays the pixels of an input image, where the intensity of each pixel is proportional to the magnitude of the gradient of the model's output with respect to that pixel, indicating its influence.
Tree Depth Plot
Histogram or bar chart showing the distribution of leaf depths in a tree-based model (such as Random Forest or XGBoost), providing insight into the model's complexity and structure.
Decision Path Visualization
Graphical representation of the specific path taken by an observation through the nodes of a decision tree or forest of trees to reach its final prediction.
Node Contribution Plot
For gradient boosting models, a visualization that breaks down the final prediction into contributions from each leaf of each tree, showing how each node adjusts the prediction.