{"id":1061,"date":"2023-01-27T12:18:55","date_gmt":"2023-01-27T11:18:55","guid":{"rendered":"https:\/\/lorentzen.ch\/?p=1061"},"modified":"2023-01-27T12:18:55","modified_gmt":"2023-01-27T11:18:55","slug":"shap-xgboost-tidymodels-love","status":"publish","type":"post","link":"https:\/\/lorentzen.ch\/index.php\/2023\/01\/27\/shap-xgboost-tidymodels-love\/","title":{"rendered":"SHAP + XGBoost + Tidymodels = LOVE"},"content":{"rendered":"\n

In this recent post<\/a>, we have explained how to use Kernel SHAP for interpreting complex linear models. As plotting backend, we used our fresh CRAN package “shapviz<\/a>“. <\/p>\n\n\n\n

“shapviz” has direct connectors to a couple of packages such as XGBoost, LightGBM, H2O, kernelshap, and more. Multiple times people asked me how to combine shapviz when the XGBoost model was fitted with Tidymodels<\/strong>. The workflow was not 100% clear to me as well, but the answer is actually very simple, thanks to Julia’s post<\/a> where the plots were made with SHAPforxgboost, another cool package for visualization of SHAP values.<\/p>\n\n\n\n

Example with shiny diamonds<\/h2>\n\n\n\n

Step 1: Preprocessing<\/h3>\n\n\n\n

We first write the data preprocessing recipe and apply it to the data rows that we want to explain. In our case, its 1000 randomly sampled diamonds. <\/p>\n\n\n

\n\t\t\t
\n\t\t\t\t
\n\t\t\t\t
R<\/div>\n\t\t\t<\/div><\/div>\n\t\t\t<\/div>\n\t\t\t
\n\n
library(tidyverse)\nlibrary(tidymodels)\nlibrary(shapviz)\n\n# Integer encode factors\ndia_recipe <- diamonds %>%\n  recipe(price ~ carat + cut + clarity + color) %>% \n  step_integer(all_nominal())\n\n# Will explain THIS dataset later\nset.seed(2)\ndia_small <- diamonds[sample(nrow(diamonds), 1000), ]\ndia_small_prep <- bake(\n  prep(dia_recipe), \n  has_role(\"predictor\"),\n  new_data = dia_small, \n  composition = \"matrix\"\n)\nhead(dia_small_prep)\n\n#     carat cut clarity color\n#[1,]  0.57   5       4     4\n#[2,]  1.01   5       2     1\n#[3,]  0.45   1       4     3\n#[4,]  1.04   4       6     5\n#[5,]  0.90   3       6     4\n#[6,]  1.20   3       4     6\n<\/pre><\/div>\n\n<\/div><\/div>\n\t\t<\/div>\n\n\n
Step 2: Fit Model<\/h3>\n\n\n\n
The next step is to tune and build the model. For simplicity, we skipped the tuning part. Bad, bad \ud83d\ude42<\/p>\n\n\n
\n\t\t\t
\n\t\t\t\t
\n\t\t\t\t
R<\/div>\n\t\t\t<\/div><\/div>\n\t\t\t<\/div>\n\t\t\t
\n\n
# Just for illustration - in practice needs tuning!\nxgboost_model <- boost_tree(\n  mode = \"regression\",\n  trees = 200,\n  tree_depth = 5,\n  learn_rate = 0.05,\n  engine = \"xgboost\"\n)\n\ndia_wf <- workflow() %>%\n  add_recipe(dia_recipe) %>%\n  add_model(xgboost_model)\n\nfit <- dia_wf %>%\n  fit(diamonds)<\/pre><\/div>\n\n<\/div><\/div>\n\t\t<\/div>\n\n\n
Step 3: SHAP Analysis<\/h3>\n\n\n\n
We now need to call shapviz()<\/code> on the fitted model. In order to have neat interpretations with the original factor labels, we not only pass the prediction data prepared in Step 1 via bake()<\/code>, but also the original data structure.<\/p>\n\n\n
\n\t\t\t
\n\t\t\t\t
\n\t\t\t\t
R<\/div>\n\t\t\t<\/div><\/div>\n\t\t\t<\/div>\n\t\t\t
\n\n
shap <- shapviz(extract_fit_engine(fit), X_pred = dia_small_prep, X = dia_small)\n\nsv_importance(shap, kind = \"both\", show_numbers = TRUE)\nsv_dependence(shap, \"carat\", color_var = \"auto\")\nsv_dependence(shap, \"clarity\", color_var = \"auto\")\nsv_force(shap, row_id = 1)\nsv_waterfall(shap, row_id = 1)\n<\/pre><\/div>\n\n<\/div><\/div>\n\t\t<\/div>\n\n\n
\"\"
Variable importance plot overlaid with SHAP summary beeswarms<\/figcaption><\/figure>\n\n\n\n
\"\"
Dependence plot for carat. Note that clarity is shown with original labels, not only integers.<\/figcaption><\/figure>\n\n\n\n
\"\"
Dependence plot for clarity. Note again that the x-scale uses the original factor levels, not the integer encoded values.<\/figcaption><\/figure>\n\n\n\n
\"\"
Force plot of the first observation<\/figcaption><\/figure>\n\n\n\n
\"\"
Waterfall plot for the first observation<\/figcaption><\/figure>\n\n\n\n
Summary<\/h3>\n\n\n\n
Making SHAP analyses with XGBoost Tidymodels is super easy.<\/p>\n\n\n\n
The complete R script can be found here<\/a>.<\/p>\n\n\n\n
<\/p>\n","protected":false},"excerpt":{"rendered":"
tidymodels and shapviz to explain XGBoost models<\/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-1061","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\/1061","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=1061"}],"version-history":[{"count":4,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/posts\/1061\/revisions"}],"predecessor-version":[{"id":1072,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/posts\/1061\/revisions\/1072"}],"wp:attachment":[{"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/media?parent=1061"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/categories?post=1061"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lorentzen.ch\/index.php\/wp-json\/wp\/v2\/tags?post=1061"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}