| Title: | Projection Pursuit Regression Tree Visualization |
|---|---|
| Description: | It was developed as a tool for exploring 'PPTreereg' (Projection Pursuit TREE of REGression). It uses various projection pursuit indexes and 'XAI' (eXplainable Artificial Intelligence) methods to help understand the model by finding connections between the input variables and prediction values of the model. The 'KernelSHAP' (Aas, Jullum and Løland (2019) <arXiv:1903.10464>) algorithm was modified to fit ‘PPTreereg’, and some codes were modified from the 'shapr' package (Sellereite, Nikolai, and Martin Jullum (2020) <doi:10.21105/joss.02027>). The implemented methods help to explore the model at the single instance level as well as at the whole dataset level. Users can compare with other machine learning models by applying it to the 'DALEX' package of 'R'. |
| Authors: | Eun-Kyung Lee [aut, ctb], HyunSun Cho [aut, cre], Nikolai Sellereite [ctb, cph] (Author of included shapr fragments), Martin Jullum [ctb, cph] (Author of included shapr fragments), Annabelle Redelmeier [ctb, cph] (Author of included shapr fragments), Norsk Regnesentral [cph] |
| Maintainer: | HyunSun Cho <[email protected]> |
| License: | GPL-3 |
| Version: | 2.0.5 |
| Built: | 2024-10-12 04:27:13 UTC |
| Source: | https://github.com/sunsmiling/pptreeregviz |
The dataXY dataset is simulated data for running Projection Pursuit Regression Tree Model.
data(dataXY)data(dataXY)
A data frame with 100 rows and 4 variables.
It contains 100 rows and 4 variables.
decision plot for PPKernelSHAP
decisionplot( PPTreeregOBJ, testObs, final.rule = 5, method = "simple", varImp = "shapImp", final.leaf = NULL, Yrange = FALSE )decisionplot( PPTreeregOBJ, testObs, final.rule = 5, method = "simple", varImp = "shapImp", final.leaf = NULL, Yrange = FALSE )
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
testObs |
test data observation |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
method |
simple or empirical method to calculate |
varImp |
|
final.leaf |
location of final leaf |
Yrange |
show the entire final prediction range of the dependent variable. Default value is FALSE. |
Decision plots are mainly used to explain individual predictions that how the model makes decision,
by focusing more on how model’s predictions reach to their expected y value with PPKernelSHAP values.
An object of the class ggplot
data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) decisionplot(Model, testX, final.rule =5, method="simple")data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) decisionplot(Model, testX, final.rule =5, method="simple")
PPTreeregObj for DALEX packageCreate Model Explainer for PPTreereg
explain_PP(PPTreeregOBJ, data, y, final.rule,...)explain_PP(PPTreeregOBJ, data, y, final.rule,...)
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
data |
data.frame or matrix - data that was used for fitting. If not provided then will be extracted from the model. Data should be passed without target column (this shall be provided as the y argument). |
y |
numeric vector with outputs / scores. If provided then it shall have the same size as data |
final.rule |
rule to calculate the final node value |
... |
arguments to be passed to methods |
This function creates a unified representation explain of PPTreereg model for cooperate with DALEX package.
An object of the class explainer.
Explanatory Model Analysis. Explore, Explain and Examine Predictive Models. https://ema.drwhy.ai/
library("DALEX") library("dplyr") data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) new_explainer <- explain_PP(Model, data = dataXY[,-1],y = dataXY[,1],final.rule= 5) DALEX::model_performance(new_explainer) %>% plot(geom = "ecdf")library("DALEX") library("dplyr") data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) new_explainer <- explain_PP(Model, data = dataXY[,-1],y = dataXY[,1],final.rule= 5) DALEX::model_performance(new_explainer) %>% plot(geom = "ecdf")
The original source for much of this came from 'shapr' package code in github.com/NorskRegnesentral/shapr/blob/master/R/features.R
feature_exact(m, weight_zero_m = 10^6)feature_exact(m, weight_zero_m = 10^6)
m |
List. Contains vector of integers indicating the feature numbers for the different groups. |
weight_zero_m |
weight_zero_m |
Below is the original license statement for 'shapr' package.
MIT License Copyright (c) 2019 Norsk Regnesentral Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
A data.table with all feature group combinations, shapley weights etc.
Nikolai Sellereite
The shapr package developed by
Nikolai Sellereite, Martin Jullum, Annabelle Redelmeier, Norsk Regnesentral.
doi:10.1016/j.artint.2021.103502 and modified some codes at
https://github.com/NorskRegnesentral/shapr
Dataset insurance is a part of dataset imported from
insurance.csv in Kaggle "Medical Cost Personal Dataset".
This data source material comes from Machine Learning with R by Brett Lantz book.
It is simply come cleaned up and, it contains 1338 rows and 7 variables. These are:
data(insurance)data(insurance)
a data frame with 1338 rows and 7 columns.
charges - Individual medical costs billed by health insurance.
age - age of primary beneficiary.
sex - insurance contractor gender, female, male.
bmi - Body mass index, providing an understanding of body, weights that are relatively high or low relative to height, objective index of body weight (kg / m ^ 2) using the ratio of height to weight, ideally 18.5 to 24.9.
children - Number of children covered by health insurance / Number of dependents.
smoker - Smoking.
region - the beneficiary's residential area in the US, northeast, southeast, southwest, northwest.
Source: https://www.kaggle.com/mirichoi0218/insurance
The insurance.csv dataset was downloaded from the Kaggle site.
The dataset was obtained from
https://www.kaggle.com/mirichoi0218/insurance on May 11, 2021.
PPTreereg
Visualize importance measure of trained PPTreereg model.
## S3 method for class 'PPimportance' plot(x, marginal = FALSE, num_var = 5, ...)## S3 method for class 'PPimportance' plot(x, marginal = FALSE, num_var = 5, ...)
x |
an importance object of the class |
marginal |
plot global importance. Default value is FALSE. |
num_var |
number of variables to show. |
... |
arguments to be passed to methods |
To visualize the variable importance values of PPTreereg model, two types of plots are
provided - importance of variables for each final node and global variable importance.
An object of the class ggplot
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) Tree.Imp <- PPimportance(Model) plot(Tree.Imp) plot(Tree.Imp, marginal = TRUE)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) Tree.Imp <- PPimportance(Model) plot(Tree.Imp) plot(Tree.Imp, marginal = TRUE)
projection pursuit regression tree plot
## S3 method for class 'PPTreereg' plot(x, font.size = 17, width.size = 1, ...)## S3 method for class 'PPTreereg' plot(x, font.size = 17, width.size = 1, ...)
x |
PPTreereg class object |
font.size |
font size of plot |
width.size |
size of eclipse in each node. |
... |
arguments to be passed to methods |
Draw projection pursuit regression tree with tree structure. It is modified from a function in party library.
plot object
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) plot(Model)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) plot(Model)
projection pursuit regression tree plot with independent variable
pp_ggparty(PPTreeregOBJ,ind_variable,final.rule=5,Rule=1, ...)pp_ggparty(PPTreeregOBJ,ind_variable,final.rule=5,Rule=1, ...)
PPTreeregOBJ |
PPTreereg class object |
ind_variable |
independent variable to show |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
Rule |
split rule 1: mean of two group means 2: weighted mean of two group means - weight with group size 3: weighted mean of two group means - weight with group sd 4: weighted mean of two group means - weight with group se 5: mean of two group medians 6: weighted mean of two group medians - weight with group size 7: weighted mean of two group median - weight with group IQR 8: weighted mean of two group median - weight with group IQR and group size |
... |
arguments to be passed to methods |
Draw projection pursuit regression tree with independent variable. It is modified
from a function in partykit library.
An object of the class ggplot
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) pp_ggparty(Model, "X1", final.rule=5)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) pp_ggparty(Model, "X1", final.rule=5)
Calculate the importance of variables in the PPTreereg model.
For local importance, weighted sum of projection coefficients with the number of data corresponding to
each node as the weighted value in each node is used.
The global importance is absolute sum of local importance.
PPimportance(PPTreeregOBJ,...)PPimportance(PPTreeregOBJ,...)
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
... |
arguments to be passed to methods |
An object of the class PPimpobj
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) PPimportance(Model)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) PPimportance(Model)
Visualize node in projection pursuit regression tree.
PPregNodeViz(PPTreeregOBJ,node.id,Rule=5)PPregNodeViz(PPTreeregOBJ,node.id,Rule=5)
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
node.id |
node ID of inner or final node |
Rule |
split rule 1: mean of two group means 2: weighted mean of two group means - weight with group size 3: weighted mean of two group means - weight with group sd 4: weighted mean of two group means - weight with group se 5: mean of two group medians 6: weighted mean of two group medians - weight with group size 7: weighted mean of two group median - weight with group IQR 8: weighted mean of two group median - weight with group IQR and group size |
This function is developed for the visualization of inner and final nodes. Visual representation of the projection coefficient value of each node and the result of projected data help understand growth process of the projection pursuit regression tree. For the inner node, two plots are provided - the bar chart style plot with projection pursuit coefficients of each variable, the histogram of the projected data. For the final node, scatter plot of observed Y vs. fitted Y according to the final rules.
An object of the class ggplot
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) PPregNodeViz(Model,node.id=1) PPregNodeViz(Model,node.id=4)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) PPregNodeViz(Model,node.id=1) PPregNodeViz(Model,node.id=4)
This function is developed to see the influence of independent variables on the range of dependent variable.
PPregVarViz(PPTreeregOBJ,var.id,indiv=FALSE, DEPTH=NULL,smoothMethod="auto", var.factor=FALSE)PPregVarViz(PPTreeregOBJ,var.id,indiv=FALSE, DEPTH=NULL,smoothMethod="auto", var.factor=FALSE)
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
var.id |
independent variable name |
indiv |
TRUE: individual group plot, FALSE: combined one plot |
DEPTH |
depth for exploration |
smoothMethod |
method in geom_smooth function |
var.factor |
TRUE when indepedent variable is a categorical variable (as factor) |
An object of the class ggplot
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) PPregVarViz(Model,"X1") PPregVarViz(Model,"X1",indiv = TRUE)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) PPregVarViz(Model,"X1") PPregVarViz(Model,"X1",indiv = TRUE)
Dependency plot using PPKernelSHAP
PPshapdependence(data_long, x, y=NULL, color_feature=NULL, smooth=TRUE)PPshapdependence(data_long, x, y=NULL, color_feature=NULL, smooth=TRUE)
data_long |
|
x |
the independent variable to see |
y |
the interaction effect by putting the values of the independent variables in different colors. |
color_feature |
display other variables with color. Default value is NULL. |
smooth |
geom_smooth option. Default value is TRUE. |
Dependency plots are designed to show the effect of one independent variable on the model's prediction.
Each point corresponds to each row of the training data,
and the y axis corresponds the PPKernelSHAP value of the variable,
indicating how much knowing the value of the variable changes the output of the model
for the prediction of the data.
An object of the class ggplot
data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long <- ppshapr_prep(Model, final.rule =5, method="simple") PPshapdependence(shap_long,x = "X1")data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long <- ppshapr_prep(Model, final.rule =5, method="simple") PPshapdependence(shap_long,x = "X1")
PPKernelSHAP for all train data setAll train data set to calculate PPKernelSHAP
ppshapr_prep(PPTreeregOBJ = NULL, final.rule = 5, method = "simple")ppshapr_prep(PPTreeregOBJ = NULL, final.rule = 5, method = "simple")
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
method |
simple or empirical method to calculate |
ppshapr_prep class object
data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long <- ppshapr_prep(Model, final.rule =5, method="simple")data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long <- ppshapr_prep(Model, final.rule =5, method="simple")
PPKernelSHAP values with empirical methodsThis function should only be called internally, and not be used as a stand-alone function. The original source for much of this came from 'shapr' package code in github.com/NorskRegnesentral/shapr/blob/master/R/predictions.R
ppshapr.empirical(PPTreeregOBJ, testObs, final.rule, final.leaf = NULL)ppshapr.empirical(PPTreeregOBJ, testObs, final.rule, final.leaf = NULL)
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
testObs |
test data observation |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
final.leaf |
location of final leaf |
Below is the original license statement for 'shapr' package.
MIT License Copyright (c) 2019 Norsk Regnesentral Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
List of empirical methods and model values
PPKernelSHAP values with simple methodsThis function should only be called internally, and not be used as a stand-alone function. The original source for much of this came from 'shapr' package code in github.com/NorskRegnesentral/shapr/blob/master/R/predictions.R
ppshapr.simple(PPTreeregOBJ, testObs, final.rule, final.leaf = NULL)ppshapr.simple(PPTreeregOBJ, testObs, final.rule, final.leaf = NULL)
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
testObs |
test data observation |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
final.leaf |
location of final leaf |
Below is the original license statement for 'shapr' package.
MIT License Copyright (c) 2019 Norsk Regnesentral Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
List of simple methods and model values
Summary plot using PPKernelSHAP
PPshapsummary(data_long,...)PPshapsummary(data_long,...)
data_long |
|
... |
arguments to be passed to methods |
A summary plot is used to see the aspects of important variables for each final node. The summary plot summarizes information about the independent variables that contributed the most to the model's prediction in the training data in the form of a density plot.
An object of the class ggplot
data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long <- ppshapr_prep(Model, final.rule =5, method="simple") PPshapsummary(shap_long)data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long <- ppshapr_prep(Model, final.rule =5, method="simple") PPshapsummary(shap_long)
Find regression tree structure using various projection pursuit indices in each split.
PPTreereg(formula,data,DEPTH=NULL,Rr=1,PPmethod="LDA", weight=TRUE,lambda=0.1,r=1,TOL.CV=0.1,selP=NULL, energy=0,maxiter=500, standardized=TRUE,even=TRUE,space=0, maxFinalNode=20,maxNodeN=10,...)PPTreereg(formula,data,DEPTH=NULL,Rr=1,PPmethod="LDA", weight=TRUE,lambda=0.1,r=1,TOL.CV=0.1,selP=NULL, energy=0,maxiter=500, standardized=TRUE,even=TRUE,space=0, maxFinalNode=20,maxNodeN=10,...)
formula |
an object of class "formula" |
data |
data frame |
DEPTH |
depth of the projection pursuit regression tree |
Rr |
cutoff rule in each node |
PPmethod |
method for projection pursuit; |
weight |
weight flag in |
lambda |
lambda in PDA index |
r |
r in Lr index |
TOL.CV |
CV limit for the final node |
selP |
number of variables for the final node in Method 5 |
energy |
energy parameter |
maxiter |
number of maximum iteration |
standardized |
standardize each X variable before fitting the tree structure. Default value is TRUE |
even |
divide evenly at each node. Default value is TRUE |
space |
space between two groups of dependent variable |
maxFinalNode |
maximum number of final node |
maxNodeN |
maximum number of observations in the final node |
... |
arguments to be passed to methods |
Tree.result projection pursuit regression tree result with
PPtreeclass object format
MSE mean squared error of the final tree
mean.G means of the observations in the final node
sd.G standard deviations of the observations in the final node.
coef.G regression coefficients for Method 3, 4 and 5
origY original dependent variable vector
origX.mean mean of original X
origX.sd standard deviation of original X
class.origX.mean means of the each independent variables in the final node
...
data(mtcars) Tree.result <- PPTreereg(mpg~.,mtcars,DEPTH=2,PPmethod="LDA") Tree.resultdata(mtcars) Tree.result <- PPTreereg(mpg~.,mtcars,DEPTH=2,PPmethod="LDA") Tree.result
PPTreereg
predict projection pursuit regression tree
## S3 method for class 'PPTreereg' predict( object, newdata = NULL, Rule = 1, final.rule = 1, classinfo = FALSE, ... )## S3 method for class 'PPTreereg' predict( object, newdata = NULL, Rule = 1, final.rule = 1, classinfo = FALSE, ... )
object |
a fitted object of class inheriting from |
newdata |
the test data set |
Rule |
split rule 1: mean of two group means 2: weighted mean of two group means - weight with group size 3: weighted mean of two group means - weight with group sd 4: weighted mean of two group means - weight with group se 5: mean of two group medians 6: weighted mean of two group medians - weight with group size 7: weighted mean of two group median - weight with group IQR 8: weighted mean of two group median - weight with group IQR and group size 9: cutoff that minimize error rates in each node |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
classinfo |
return final node information. Default value is FALSE |
... |
arguments to be passed to methods |
Predict class for the test set with the fitted projection pursuit regression tree and calculate prediction error.
Numeric
data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) predict(Model)data(dataXY) Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) predict(Model)
Print PP.Tree.reg result
## S3 method for class 'PPTreereg' print( x, tree.print = TRUE, coef.print = FALSE, cutoff.print = FALSE, verbose = TRUE, final.rule = 1, ... )## S3 method for class 'PPTreereg' print( x, tree.print = TRUE, coef.print = FALSE, cutoff.print = FALSE, verbose = TRUE, final.rule = 1, ... )
x |
PPTreereg object |
tree.print |
print the tree structure when TRUE |
coef.print |
print the projection coefficient in each node when TRUE |
cutoff.print |
print the cutoff values in each node when TRUE |
verbose |
print if TRUE, no output if FALSE |
final.rule |
rule to calculate the final node value |
... |
arguments to be passed to methods |
Print the projection pursuit regression tree result
tree print
The original source for much of this came from 'shapr' package code in github.com/NorskRegnesentral/shapr/blob/master/R/shapley.R Below is the original license statement for 'shapr' package.
shapley_weights(m, N, n_components, weight_zero_m = 10^6)shapley_weights(m, N, n_components, weight_zero_m = 10^6)
m |
m |
N |
N |
n_components |
n_components |
weight_zero_m |
weight_zero_m |
MIT License Copyright (c) 2019 Norsk Regnesentral Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Numeric
Nikolai Sellereite
The shapr package developed by
Nikolai Sellereite, Martin Jullum, Annabelle Redelmeier, Norsk Regnesentral.
doi:10.1016/j.artint.2021.103502 and modified some codes at
https://github.com/NorskRegnesentral/shapr
submodular pick algorithm PP SP-LIME
Pick several data containing various information for each final node for PPTreereg
submodular Pick (SP-LIME) was developed (Ribeiro et al., 2016) to selects
representative data with important information to determine the
reliability of model based on the LIME algorithm.
In order to extract data for each final node in the PPTreereg model,
PP SP-LIME was proposed based on SP-LIME.
subpick(data_long, final.leaf, obsnum = 5)subpick(data_long, final.leaf, obsnum = 5)
data_long |
|
final.leaf |
location of final leaf |
obsnum |
The number of budgets (instance to be selected). Default value is 1. |
Observation names and their original values as data
Ribeiro, Marco Tulio, Sameer Singh, and Carlos Guestrin. "" Why should i trust you?" Explaining the predictions of any classifier." Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. 2016. doi:10.1145/2939672.2939778 https://github.com/marcotcr/lime/blob/master/lime/submodular_pick.py
data("dataXY") Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long=ppshapr_prep(Model,final.rule =3,method="simple") subpick(shap_long,final.leaf = 1, obsnum = 5)data("dataXY") Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) shap_long=ppshapr_prep(Model,final.rule =3,method="simple") subpick(shap_long,final.leaf = 1, obsnum = 5)
PPTreereg resultsummary PPTreereg result
## S3 method for class 'PPTreereg' summary(object, c = NA, ...)## S3 method for class 'PPTreereg' summary(object, c = NA, ...)
object |
a fitted object of class inheriting from |
c |
choose node id to summary. Default value is FALSE. |
... |
arguments to be passed to methods |
summary the projection pursuit regression tree result
coefficient results of tree
waterfall plot for PPKernelSHAP
waterfallplot( PPTreeregOBJ, testObs, final.rule = 5, method = "simple", final.leaf = NULL )waterfallplot( PPTreeregOBJ, testObs, final.rule = 5, method = "simple", final.leaf = NULL )
PPTreeregOBJ |
PPTreereg class object - a model to be explained |
testObs |
test data observation |
final.rule |
final rule to assign numerical values in the final nodes. 1: mean value in the final nodes 2: median value in the final nodes 3: using optimal projection 4: using all independent variables 5: using several significant independent variables |
method |
simple or empirical method to calculate |
final.leaf |
location of final leaf |
Waterfall plot is mainly used to explain individual predictions,
and is suitable for showing an explanation when a single piece of data is
entered as an input using PPKernelSHAP values.
An object of the class ggplot
data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) waterfallplot(Model, testX, final.rule =5, method="simple")data(dataXY) testX <- dataXY[1,-1] Model <- PPTreereg(Y~., data = dataXY, DEPTH = 2) waterfallplot(Model, testX, final.rule =5, method="simple")
The original source for much of this came from 'shapr' package code in github.com/NorskRegnesentral/shapr/blob/master/R/shapley.R Below is the original license statement for 'shapr' package.
weight_matrix(X, normalize_W_weights = TRUE)weight_matrix(X, normalize_W_weights = TRUE)
X |
X |
normalize_W_weights |
default is TRUE |
MIT License Copyright (c) 2019 Norsk Regnesentral Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Numeric matrix
Nikolai Sellereite
The shapr package developed by
Nikolai Sellereite, Martin Jullum, Annabelle Redelmeier, Norsk Regnesentral.
doi:10.1016/j.artint.2021.103502 and modified some codes at
https://github.com/NorskRegnesentral/shapr