XAI: Explainable AI

Name: XAI: Explainable AI
Rating: 4.3 (241 reviews)

How to explain your machine learning models in Python

Created byVioleta Misheva

Last updated 1/2022

English

EnglishKorean [Auto],

What you'll learn

How to explain machine learning models using various techniques

Course content

6 sections • 14 lectures • 2h 1m total length

What is model explainability?9:25
Explore explainable and interpretable machine learning, differentiate explanations, and learn to develop explanations for supervised models using open source Python packages.
Why explainability? And when not?5:40
Explore why explainable AI matters, balancing transparent and complex models for trust, bias detection, robustness, and domain understanding, while warning against explanations that enable gaming the system.
Explainability: what and when?

Types of explainability and taxonomy of explanations4:59
Explore why explainability matters in machine learning through contrastive, selective, and consistent explanations, review intrinsic vs post hoc methods and local vs global explanations with lime, anchors, and Shapley values.
Explainability in the model development process5:37
Define explainability needs for diverse stakeholders and integrate explainability early in the model development lifecycle. Apply multiple explanation methods as a flexible process, monitor explanations alongside model performance in production.
Types of explainability

Setting up a Python virtual environment and required packages1:00
Problem intro7:35
Learn to set up a Python data science environment, load the diabetes dataset, build a RandomForestRegressor, and explore feature importances, correlations, and permutation-based explainability methods for XAI.
On transparent models and RuleFit12:36
Explore inherently transparent models and rule-based learning with RuleFit, which generates decision rules from trees and selects key features with LASSO for interpretable predictions.
Partial dependency plots: what they are, how to apply them10:18
Explore partial dependency plots to reveal the average marginal effect of features on the target. See how age, BMI, and their interaction shape predictions and note the independence assumption.
Individual conditional expectations10:31
Explore individual conditional expectations (ICE plots) as a visual tool to reveal heterogeneous effects, compare with partial dependency plots, and center ICE curves for easier interpretation.
Apply PDP and ICE plots on a sklearn dataset

Global surrogate models5:07
Train a global surrogate on a black box's predictions to obtain a simple, transparent approximation. Use r-squared to assess fidelity with a shallow tree approximating a random forest.
Feature importances6:53
Explore feature importances and permutation importance, which measure how permuting a feature increases model error to provide global insight across models, including random forest and linear regression, via sklearn.
Global surrogate and feature importances

LIME9:47
Learn how lime provides local interpretable model-agnostic explanations for individual predictions in classification and regression. It perturbs data, trains a weighted surrogate model on a neighborhood to reveal feature importance.
Shapley values14:54
Explore how Shapley values attribute each feature's contribution to a model's prediction for local explanations. Use the shap package, TreeShap for tree models, and visualize global and instance-level explanations.
Let's go local: LIME and Shapley values

Requirements

Write basic Python code and know basics of machine learning

Description

Machine learning models are becoming more and more popular. But not every user is convinced in their utility and usability. How and when can we trust the models? If our model has rejected a loan applicant, can we explain to them why that is the case? What types of explanations about the model or its behavior can we provide? What does it even means to explain a model?

We address these and other questions in this course on Machine learning or AI explainability (also called XAI in short). We will introduce theoretical approaches and build a hands-on understanding of various explainability techniques in Python.

The course builds an overview of XAI approaches before going into details of different types of explanations: visual, explanations of the overall model behavior (so-called global), as well as of how the model reached its decision for every single prediction(so-called local explanations). We will apply each presented approach to a regression and/or classification task; and you will gain ever more practice with the techniques using the hands on assignments.

By the end of the course, you should have an understanding of the current state-of-the-art XAI approaches, their benefits and pitfalls. You will also be able to use the tools learned here in your own use cases and projects.

XAI is a rapidly developing research field with many open-ended questions. But one thing is certain: it is not going anywhere, the same way Machine learning and AI are here to stay.

Who this course is for:

Students with some machine learning and Python coding experience

XAI: Explainable AI

What you'll learn

Explore related topics

Course content

Introduction2 lectures • 15min

Types of explainability2 lectures • 11min

Visual explanations5 lectures • 42min

Global explanations2 lectures • 12min

Local explanations2 lectures • 25min

Wrap up1 lecture • 3min

Requirements

Description

Who this course is for: