| Title: | Visualization of Subgroups for Decision Trees |
|---|---|
| Description: | Provides a visualization for characterizing subgroups defined by a decision tree structure. The visualization simplifies the ability to interpret individual pathways to subgroups; each sub-plot describes the distribution of observations within individual terminal nodes and percentile ranges for the associated inner nodes. |
| Authors: | Ashwini Venkatasubramaniam [aut, cre], Julian Wolfson [aut, ctb] |
| Maintainer: | Ashwini Venkatasubramaniam <[email protected]> |
| License: | GPL-3 |
| Version: | 0.8.1 |
| Built: | 2024-11-05 03:57:03 UTC |
| Source: | https://github.com/ashwinikv/vistree |
The variables are as follows:
data(blsdata)data(blsdata)
A data frame with 226 rows and 26 variables
trt. Treatment
sex. Sex
bmi0. BMI
snackkcal0. Snacking kilo calories
srvgfv0. Serving size of fruits and vegetables
srvgssb0. Serving size of beverages
kcal24h0.
edeq01.
edeq02.
edeq13.
edeq14.
edeq15.
edeq22.
edeq23.
edeq25.
edeq26.
cdrsbody0. Body image
weighfreq0. Weighing frequency
freqff0. Fast food frequency
age. Age
tfactor1.
tfactor2.
tfactor3.
mlhfbias0.
fwahfbias0.
rrvfood. Relative reinforcement of food
data(blsdata)data(blsdata)
Decision tree structure
l_node(newtree, node_id = 1, start_criteria = character(0))l_node(newtree, node_id = 1, start_criteria = character(0))
newtree |
Decision tree generated as a party object |
node_id |
Node ID |
start_criteria |
Character vector |
Function to adjust the transparency and define the color scheme within the visualization.
makeTransparent(colortype, alpha)makeTransparent(colortype, alpha)
colortype |
Color palette |
alpha |
Transparency |
Identifies splits and relevant criteria
minmax_mat(str, varnms, Y, interval)minmax_mat(str, varnms, Y, interval)
str |
Structure of pathway from the root node in the decision tree to each terminal node |
varnms |
Names of covariates |
Y |
Response variable in the dataset |
interval |
logical. Continuous response (interval = FALSE) and Categorical response (interval = TRUE). |
Generates the pathway from the root node to individual terminal nodes of a decision tree generated as a party object using the partykit package.
path_node(newtree, idnumber = 0)path_node(newtree, idnumber = 0)
newtree |
Decision tree generated as a party object |
idnumber |
Terminal ID number |
This function is utilized to generate a series of sub-plots, where each subplot corresponds to individual terminal nodes within the decision tree structure. Each subplot is composed of a histogram (or a barchart) that displays the distribution for the relevant subgroup and colored horizontal bars that summarize the set of covariate splits.
plot_minmax(My, X, Y, str, color.type, alpha, add.p.axis, add.h.axis, cond.tree, text.main, text.bar, text.round, text.percentile, density.line, text.title, text.axis, text.label)plot_minmax(My, X, Y, str, color.type, alpha, add.p.axis, add.h.axis, cond.tree, text.main, text.bar, text.round, text.percentile, density.line, text.title, text.axis, text.label)
My |
A matrix to define the split points within the decision tree structure |
X |
Covariates |
Y |
Response variable |
str |
Structure of pathway from the root node in the decision tree to each terminal node |
color.type |
Color palettes. (rainbow_hcl = 1; heat_hcl = 2; terrain_hcl = 3; sequential_hcl = 4; diverge_hcl = 5) |
alpha |
Transparency of individual horizontal bars. Choose values between 0 to 1. |
add.p.axis |
logical. Add axis for the percentiles (add.p.axis = TRUE), remove axis for the percentiles (add.p.axis = FALSE). |
add.h.axis |
logical. Add axis for the outcome (add.h.axis = TRUE), remove axis for the outcome (add.h.axis = FALSE). |
cond.tree |
Tree as a party object |
text.main |
Change the size of the main titles |
text.bar |
Change the size of the text in the horizontal bar and below the bar plot |
text.round |
Round the threshold displayed on the bar |
text.percentile |
Change the size of the percentile title |
density.line |
Draw a density line |
text.title |
Change the size of the text in the title |
text.axis |
Change the size of the text of axis labels |
text.label |
Change the size of the axis annotation |
Identifies the splitting criteria for the relevant node leading to lower level inner nodes or a terminal node.
ptree_criteria(newtree, node_id, left)ptree_criteria(newtree, node_id, left)
newtree |
Decision tree |
node_id |
Node id |
left |
Splits to the left |
Identifies a node that corresponds to the left split
ptree_left(newtree, start_id)ptree_left(newtree, start_id)
newtree |
Decision tree generated as a party object |
start_id |
Character vector |
Identifies a node that corresponds to the right split
ptree_right(newtree, start_id)ptree_right(newtree, start_id)
newtree |
Decision tree generated as a party object |
start_id |
Character vector |
Identifies the predicted outcome value for the relevant node.
ptree_y(newtree, node_id)ptree_y(newtree, node_id)
newtree |
Decision tree generated as a party object |
node_id |
Node ID |
Parsing function
trim(x)trim(x)
x |
String |
This visualization characterizes subgroups defined by a decision tree structure and identifies the range of covariate values associated with outcome values in each subgroup.
visTree(cond.tree, rng = NULL, interval = FALSE, color.type = 1, alpha = 0.5, add.h.axis = TRUE, add.p.axis = TRUE, text.round = 1, text.main = 1.5, text.bar = 1.5, text.title = 1.5, text.label = 1.5, text.axis = 1.5, text.percentile = 0.7, density.line = TRUE)visTree(cond.tree, rng = NULL, interval = FALSE, color.type = 1, alpha = 0.5, add.h.axis = TRUE, add.p.axis = TRUE, text.round = 1, text.main = 1.5, text.bar = 1.5, text.title = 1.5, text.label = 1.5, text.axis = 1.5, text.percentile = 0.7, density.line = TRUE)
cond.tree |
Decision tree generated as a party object. |
rng |
Restrict plotting to a particular set of nodes. Default value is set as NULL. |
interval |
logical. Continuous outcome (interval = FALSE) and Categorical outcome (interval = TRUE). |
color.type |
Color palettes (rainbow_hcl = 1; heat_hcl = 2; terrain_hcl = 3; sequential_hcl = 4 ; diverge_hcl = 5) |
alpha |
Transparency for horizontal colored bars in each subplot. Values between 0 to 1. |
add.h.axis |
logical. Add axis for the outcome distribution (add.h.axis = TRUE), remove axis for the outcome (add.h.axis = FALSE). |
add.p.axis |
logical. Add axis for the percentiles (add.p.axis = TRUE) computed over covariate values, remove axis for the percentiles (add.p.axis = FALSE). |
text.round |
Round the threshold displayed on the horizontal bar |
text.main |
Change the size of the main titles |
text.bar |
Change the size of the text in the horizontal bar |
text.title |
Change the size of the text in the title |
text.label |
Change the size of the axis annotation |
text.axis |
Change the size of the text of axis labels |
text.percentile |
Change the size of the percentile title |
density.line |
logical. Draw a density line. (density.line = TRUE). |
Ashwini Venkatasubramaniam and Julian Wolfson
data(blsdata) newblsdata<-blsdata[,c(7,21, 22,23, 24, 25, 26)] ## Continuous response ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.5) ## Repeated covariates in the splits of the decision tree ptree2<-partykit::ctree(kcal24h0~skcal+rrvfood+resteating+age, data = blsdata) visTree(ptree2, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.5) ## Categorical response blsdataedit<-blsdata[,-7] blsdataedit$bin<-0 blsdataedit$bin<-cut(blsdata$kcal24h0, unique(quantile(blsdata$kcal24h0)), include.lowest = TRUE, dig.lab = 4) names(blsdataedit)[26]<-"kcal24h0" ptree3<-partykit::ctree(kcal24h0~hunger+rrvfood+resteating+liking, data = blsdataedit) visTree(ptree3, interval = TRUE, color.type = 1, alpha = 0.6, text.percentile = 1.2, text.bar = 1.8) ## Other decision trees (e.g., rpart) ptree4<-rpart::rpart(kcal24h0~wanting+liking+rrvfood, data = newblsdata, control = rpart::rpart.control(cp = 0.029)) visTree(ptree4, text.bar = 1.8, text.label = 1.4, text.round = 1, density.line = TRUE, text.percentile = 1.3) ## Change the color scheme and transparency of the horizontal bars ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.65, color.type = 3) ## Remove the axes corresponding to the percentiles and the response values. ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.65, color.type = 3, add.p.axis = FALSE, add.h.axis = FALSE) # Remove the density line over the histograms ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.65, color.type = 3, density.line = FALSE)data(blsdata) newblsdata<-blsdata[,c(7,21, 22,23, 24, 25, 26)] ## Continuous response ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.5) ## Repeated covariates in the splits of the decision tree ptree2<-partykit::ctree(kcal24h0~skcal+rrvfood+resteating+age, data = blsdata) visTree(ptree2, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.5) ## Categorical response blsdataedit<-blsdata[,-7] blsdataedit$bin<-0 blsdataedit$bin<-cut(blsdata$kcal24h0, unique(quantile(blsdata$kcal24h0)), include.lowest = TRUE, dig.lab = 4) names(blsdataedit)[26]<-"kcal24h0" ptree3<-partykit::ctree(kcal24h0~hunger+rrvfood+resteating+liking, data = blsdataedit) visTree(ptree3, interval = TRUE, color.type = 1, alpha = 0.6, text.percentile = 1.2, text.bar = 1.8) ## Other decision trees (e.g., rpart) ptree4<-rpart::rpart(kcal24h0~wanting+liking+rrvfood, data = newblsdata, control = rpart::rpart.control(cp = 0.029)) visTree(ptree4, text.bar = 1.8, text.label = 1.4, text.round = 1, density.line = TRUE, text.percentile = 1.3) ## Change the color scheme and transparency of the horizontal bars ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.65, color.type = 3) ## Remove the axes corresponding to the percentiles and the response values. ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.65, color.type = 3, add.p.axis = FALSE, add.h.axis = FALSE) # Remove the density line over the histograms ptree1<-partykit::ctree(kcal24h0~., data = newblsdata) visTree(ptree1, text.axis = 1.3, text.label = 1.2, text.bar = 1.2, alpha = 0.65, color.type = 3, density.line = FALSE)