This post is mainly about the third approach. Its beauty is that we get information about all interactions. The downside: it is as good\/bad as partial dependence functions. And: the statistics are computationally very expensive to compute (of order n^2). <\/p>\n\n\n\n
Different R packages offer some of these H-statistics, including {iml}, {gbm}, {flashlight}, and {vivid}. They all have their limitations. This is why I wrote the new R package {hstats<\/a>}:<\/p>\n\n\n\n In Python, there is the very interesting project artemis<\/a>. I will write a post on it later. <\/p>\n\n\n\n Furthermore, a global measure of non-additivity (proportion of prediction variability unexplained by main effects), and a measure of feature importance is available. For technical details and references, check the following pdf<\/a> or github<\/a>.<\/p>\n\n\n\n Let’s fit a probability random forest on iris species.<\/p>\n\n\n Interpretation:<\/strong> <\/p>\n\n\n\n Here, we consider a random forest regression on “Sepal.Length”.<\/p>\n\n\n\n Interpretation<\/strong><\/p>\n\n\n\n The complete R script can be found here<\/a>. More examples and background can be found on the Github<\/a> page of the project.<\/p>\n\n\n\n <\/p>\n","protected":false},"excerpt":{"rendered":" What makes a ML model a black-box? It is the interactions. Without any interactions, the ML model is additive and can be exactly described. Studying interaction effects of ML models is challenging. The main XAI approaches are: This post is mainly about the third approach. Its beauty is that we get information about all interactions. […]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[16,17,9],"tags":[5],"class_list":["post-1279","post","type-post","status-publish","format-standard","hentry","category-machine-learning","category-programming","category-statistics","tag-r"],"featured_image_src":null,"author_info":{"display_name":"Michael Mayer","author_link":"https:\/\/lorentzen.ch\/index.php\/author\/michael\/"},"_links":{"self":[{"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/posts\/1279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/comments?post=1279"}],"version-history":[{"count":1,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/posts\/1279\/revisions"}],"predecessor-version":[{"id":1286,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/posts\/1279\/revisions\/1286"}],"wp:attachment":[{"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/media?parent=1279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/categories?post=1279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/tags?post=1279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}![\"\"](\"https:\/\/lorentzen.ch\/wp-content\/uploads\/2023\/08\/logo.png\"){kind=logo}
<\/figure>\n\n\n\n\n
Statistics supported by {hstats}<\/h2>\n\n\n\n
![\"\"](\"https:\/\/lorentzen.ch\/wp-content\/uploads\/2023\/08\/image-1024x288.png\")
<\/figure>\n\n\n\nClassification example<\/h2>\n\n\n\n
library(ranger)\nlibrary(ggplot2)\nlibrary(hstats)\n\nv <- setdiff(colnames(iris), \"Species\")\nfit <- ranger(Species ~ ., data = iris, probability = TRUE, seed = 1)\ns <- hstats(fit, v = v, X = iris) # 8 seconds run-time\ns\n# Proportion of prediction variability unexplained by main effects of v:\n# setosa versicolor virginica \n# 0.002705945 0.065629375 0.046742035\n\nplot(s, normalize = FALSE, squared = FALSE) +\n ggtitle(\"Unnormalized statistics\") +\n scale_fill_viridis_d(begin = 0.1, end = 0.9)\n\nice(fit, v = \"Petal.Length\", X = iris, BY = \"Petal.Width\", n_max = 150) |> \n plot(center = TRUE) +\n ggtitle(\"Centered ICE plots\")\n<\/pre><\/div>\n\n<\/div><\/div>\n\t\t<\/div>\n\n\n
![\"\"](\"https:\/\/lorentzen.ch\/wp-content\/uploads\/2023\/08\/image-1-1024x379.png\")
![\"\"](\"https:\/\/lorentzen.ch\/wp-content\/uploads\/2023\/08\/image-2-1024x385.png\")
\n
DALEX example<\/h2>\n\n\n\n
library(DALEX)\nlibrary(ranger)\nlibrary(hstats)\n\nset.seed(1)\n\nfit <- ranger(Sepal.Length ~ ., data = iris)\nex <- explain(fit, data = iris[-1], y = iris[, 1])\n\ns <- hstats(ex) # 2 seconds\ns # Non-additivity index 0.054\nplot(s)\nplot(ice(ex, v = "Sepal.Width", BY = "Petal.Width"), center = TRUE)\n<\/pre><\/div>\n\n\n\n
![\"\"](\"https:\/\/lorentzen.ch\/wp-content\/uploads\/2023\/08\/image-3.png\")
![\"\"](\"https:\/\/lorentzen.ch\/wp-content\/uploads\/2023\/08\/image-4.png\")
\n
Try it out!<\/h3>\n\n\n\n