parcats.Rmd
parcats
requires an alluvial plot created with easyalluvial
to create an interactive parrallel categories diagram.
suppressPackageStartupMessages( require(tidyverse) ) suppressPackageStartupMessages( require(easyalluvial) ) suppressPackageStartupMessages( require(parcats) )
p = alluvial_wide(mtcars2, max_variables = 5) parcats(p, marginal_histograms = TRUE, data_input = mtcars2)
Machine Learning models operate in a multidimensional space and their response is hard to visualise. Model response and partial dependency plots attempt to visualise ML models in a two dimensional space. Using alluvial plots or parrallel categories diagrams we can increase the number of dimensions.
Here we see the response of a random forest model if we vary the three variables with the highest importance while keeping all other features at their median/mode value.
df = select(mtcars2, -ids ) m = randomForest::randomForest( disp ~ ., df) imp = m$importance dspace = get_data_space(df, imp, degree = 3) pred = predict(m, newdata = dspace) p = alluvial_model_response(pred, dspace, imp, degree = 3) parcats(p, marginal_histograms = TRUE, imp = TRUE, data_input = df)