calculates a dataspace based on the modeling dataframe and the importance of the explanatory variables. It only considers the most important variables as defined by the degree parameter. It selects a number (defined by bins) of sensible single values spread over the range of the numeric variables and creates all possible value combinations among the most important variables. The values of the remaining variables are set to mode(factors) or median(numerics).
get_data_space(df, imp, degree = 4, bins = 5, max_levels = 10)dataframe, training data
dataframe, with not more then two columns one of them numeric containing importance measures and one character or factor column containing corresponding variable names as found in training data.
integer, number of top important variables to select. For plotting more than 4 will result in two many flows and the alluvial plot will not be very readable, Default: 4
integer, number of bins for numeric variables, and maximum number of levels for factor variables, increasing this number might result in too many flows, Default: 5
integer, maximum number of levels per factor variable, Default: 10
data frame
It selects a the top most important variables based on the degree
parameter and bins the numeric variables using
manip_bin_numerics, while leaving categoric
variables unchanged. The number of bins for each numeric variable is set to
bins -2. Next the median is picked for each of the bins and the min and the
max value is added for each numeric variable So that we get (median(bin) X
bins -2, max, min) for each numeric variable. Then all possible combinations
between those values and the categoric factor levels are created. The total
number of all possible combinations defines the range of the data space. The
values of the remaining variables are set to mode(factors) or
median(numerics).
this model visualisation approach follows the "visualising the model in the dataspace" principle as described in Wickham H, Cook D, Hofmann H (2015) Visualizing statistical models: Removing the blindfold. Statistical Analysis and Data Mining 8(4) <doi:10.1002/sam.11271>
df = mtcars2[, ! names(mtcars2) %in% 'ids' ]
m = randomForest::randomForest( disp ~ ., df)
imp = m$importance
dspace = get_data_space(df, imp)