Having trouble explaining machine learning models? You’re not alone. The black-box nature of some gradient boosting, forest and deep learning models are often too complex for people to understand by directly inspecting model parameters.
In this video, SAS Machine Learning Developer Funda Gunes explains various model-agnostic interpretability techniques available in SAS Viya that enable you to explain and understand machine learning models. Methods include partial dependency (PD) plots, independent conditional expectation (ICE) plots, local interpretable model-agnostic explanations (LIME), and Shapley values.
Video Highlights
0:11 – Healthcare use case
2:32 – Why machine learning models are black box
4:36 – Why model interpretability?
5:43 – Types of model interpretability methods
7:33 – Post-hoc explanations
12:24 – Partial dependence (PD) Plot
13:29 – Local interpretability
14:30 – Independent conditional expectation (ICE) plots
15:45 – Local interpretable model-agnostic explanations (LIME)
17:03 – Kernal SHAP explanation
18:14 – Accessing interpretability through programmatic usage
Read the paper
And here's a related a video tutorial, Opening the Black Box of Model Interpretability in SAS® Viya® that goes into many of the same topics, including gradient boosting:
Video Highlights
02:01 – Overview of Model Interpretability
03:54 – Gradient Boosting model
05:30 – Building a Modeling Pipeline and Exploring Interpretable Models in SAS Viya
14:23 – Calculating Partial Dependence Plots
16:57 – Exploring Individual Conditional Expectation Plots
23:51 – Local Interpretable Model-Agnostic Explanations (LIME Plot) overview
27:19 – Kernel Shapley Additive Explanations (SHAP Plot)
Related Resources
How to explain your black-box models in SAS® Viya®
Interpreting Machine Learning Models in SAS (video how-to tutorial)
Machine Learning Fundamentals (video how-to tutorial)
SAS® Visual Data Mining and Machine Learning Documentation
Explain Model Action Set
Github repository
