How do I determine query points?

Question

Mark 2024-9-3，12:07

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2149759-how-do-i-determine-query-points

评论： Mark 2024-9-4，8:12

Hi,

I've recently been introduced to Shapley values and I'm trying to follow the Matlab tutorial on these but I've unsure about the section on query Points.

In my instance, I am trying to determine the importance of different predictors from a classification tree based on a table of 19 different possible predictors. The output from this tree will give me a category of either A, B, C or D, with A being the preferential category.

In the tutorial on matlab, when determining query points, the code written is:

queryPoint = tbl(end,:)

By this, it is taking the last line of the tbl in question. This may be a stupid question, but how do I adapt this to my work? Do I replace it with something along the lines of:

queryPoint = tbl(x,:)

where x is the line number of a datapoint that is categorised as A

Apolgies if it sounds dumb, I'm very new to the world of machine learning techniques but thank you for any help?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Dheeraj 2024-9-4，5:05

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2149759-how-do-i-determine-query-points#answer_1510269

编辑：Dheeraj 2024-9-4，5:06

在 MATLAB Online 中打开

Hi Mark,

To select a query point for Shapley value analysis in your classification tree model, you can adapt the MATLAB code by choosing a specific data point (row) from your dataset that has been classified into the category of interest, such as A. For example, if you want to analyze why a particular instance was classified as A, you would replace the query point selection with the row number corresponding to that instance, like

queryPoint = tbl(x,:)

where x is the row number of the desired data point. Alternatively, you can automatically select a query point by identifying rows where the category is A using a command like

find(tbl.Category == 'A')

and then choosing one of those rows for further analysis. This allows you to tailor the Shapley value analysis to the specific cases that are most relevant to your study.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Mark 2024-9-4，8:12

Thanks Dheeraj!

I had a feeling I'd have to do it effectively manually per instance of category "A", but I suspect a combination of laziness or hope that there would be some way of scanning all the data make the query point line effectively redundant and having it done automatically.

请先登录，再进行评论。

Answer 2

Sahas 2024-9-4，7:44

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2149759-how-do-i-determine-query-points#answer_1510429

在 MATLAB Online 中打开

Hi @Mark,

As per my understanding, you would like to analyze a specific data point associated to a specific category. You are using “queryPoint” property to understand the contribution of each feature in the model.

Assuming you are using MATALB R2023b version, specify the query point by using the “QueryPoint” name-value argument instead of “QueryPoints”. This is because before MATLAB R2024a, you can only specify one query point for calculating the “Shapley values”.

Refer to the following MathWorks documentation of MATLAB R2023b version to know how to compute “Shapley values” for a query point: https://www.mathworks.com/help/releases/R2023b/stats/shapley.html

In the above cited example, you can replace the “end” keyword with the index of that particular row to analyze a specific data point in category “A”. Take a look at the following code snippet for a better understanding:

% Assume "Category" is the name of the column
% Find the index of the first occurrence of category "A"
x = find(tbl.Category == 'A', 1); 
% Select that row to analyze
queryPoint = tbl(x,:);

Follow the steps in the given in the documentation to compute ”Shapley values” with a single or multiple query points according to the requirements and the MATLAB version being used.

Refer to the following MathWorks documentation of MATLAB R2024a version to know how to compute “Shapley values” for multiple query points: https://www.mathworks.com/help/stats/shapley.html

Hope this is beneficial!