Interpretable AI Series (3) Model-agnostic methods: Global interpretability

The 3rd chapter is mainly about the first part of the model-agnostic methods: global interpretability, which interprets the influences of features on the dependent variable (label) on the whole model level. The model used in this chapter is the random forest, or decision tree ensemble, while the method can be easily extended to other black-box models. Here "model-agnostic" means these methods can be used to interpret any model without modifying the details of the method.

The first section introduced the dataset: high school student performance and several other features, such as race, parent level of education, etc, followed by an introduction of exploratory data analysis (EDA for short). Here the author mainly used histograms to show the grade distribution according to different categorical features, and their combinations.

The following section recaps tree ensembles, including 2 possible approaches of an ensemble: bagging (random forest) and boosting (adaptive boosting and gradient boosting). How to train a random forest, adaptive boost, and gradient boosting trees with Scikit-learn, followed by how to use precision, recall, and f1 score to evaluate the performance of a model. All these provided a necessary background for the topic of this chapter.

The third section is a brief introduction to the random-forest-specific interpretation technique, a weighted sum of the feature importance of each element (tree), and the corresponding implementation in Python.

The last section is the essence of this chapter, which consists of 2 methods: partial dependence plots (PDP for short), and the method how investigating the interaction between features.

A package named "pdpbox" is used to plot the partial dependence plots. The author plotted PDP on the student performance dataset and interpreted the plot thoroughly. While PDP provides many insights about the influences of features, it assumes features are independent of each other, which is not true in most cases. In others words, if there're interactions between features, the PDP won't be trusted completely. Here comes the feature interaction plot. Like a heatmap, the x- and y-axis are values of that two categorical interactive features. For example, different races on the x-axis, and different parent education levels on the y-axis. The color of an intersected block represents the influence of the combination of the two categorical values.

Comments

Post a Comment

Popular posts from this blog

2023: On the Road

There're so many dramas happened in the past 3 years. People say there'll be more in the coming 2023. Basically, I agree with this. But no matter what happens in the physical world, we need to concentrate on something really important and interesting. In the past 2022, my most profound change was, that I started thinking seriously about the future living style. My family and friends mostly live like plants, fixed by our roots, and change ourselves to adapt to the environment. We are where we are, proud of its history and culture, and contained by it. It'll not be the case anymore. In 2022, I learned, read, listened, wrote, and spoke in second languages, especially English and Japanese. They bring me a new world. I also read several books on Chinese and middle east histories. The world before me is fusing as a whole sphere. China and Chinese people are a part, and the most familiar part for me, the starting point of my exploration. It gives me enough confidence as a foreig...
Read more

Yet another advice to kids

Sharon posted some advice to the youth (although she herself is young). She suggested being independent, with no fear of failure, which is excellent advice. I'd like to give some more "concrete" tips to kids or myself in childhood (if I have the chance to tell that me). First things first, the following suggestions can't make you more competitive than your peers or perform better in exams. or win the girl make you fascinated. They make your everyday life more exciting and fancy even if you're not rich or powerful. They're neither expensive (so you can achieve these goals even if not born into a rich family) nor painful. What you need is to know them, and keep doing them. OK, here we go. 1. Find a sport, enjoy it, and be good at it. Sport is a way to talk with your body, with the physical world, express yourself, and communicate with people. 2. Be honest, especially to yourself. Living honestly in Chinese society looks stupid and clumsy. Never mind. Being hones...
Read more

The Joy of Reading in Natural Light

Nowadays I read books in paperbacks much less than ebooks. Not because I don't like reading paper books, but after reading for so many years on screen, my eyes are very sensitive to the brightness of reading materials. It's very hard for me to read characters when the material is not bright enough. Unfortunately, the light in my living room is not bright enough for me. This morning it was cloudy, while in the afternoon it got clear, and the blue sky and white clouds appeared. So I moved the desk near the window. To my surprise, the characters on paper books are much more clear and more comfortable to read than in artificial light. Maybe such a simple fact is not worthy to mention. But to me, a programmer spent too much time with the keyboard and screen, this reminds me of the memories of childhood, reading a book, or writing homework in the yard. The characters blurred at sunset. That's when we were going to have super. The window of my living room is east-forwarding, so to...
Read more