| Title: | Plots for Model Sensitivity and Variable Importance |
| Version: | 0.1.3 |
| Description: | Draws tornado plots for model sensitivity to univariate changes. Implements methods for many modeling methods including linear models, generalized linear models, survival regression models, and arbitrary machine learning models in the caret package. Also draws variable importance plots. |
| License: | GPL-3 |
| Encoding: | UTF-8 |
| Suggests: | testthat, caret, glmnet, randomForest, knitr, rmarkdown |
| RoxygenNote: | 7.3.0 |
| Imports: | survival, assertthat, ggplot2, scales, grid, gridExtra, rlang |
| VignetteBuilder: | knitr |
| URL: | https://github.com/bertcarnell/tornado |
| BugReports: | https://github.com/bertcarnell/tornado/issues |
| NeedsCompilation: | no |
| Packaged: | 2024-01-20 02:33:21 UTC; bertc |
| Author: | Rob Carnell [aut, cre] |
| Maintainer: | Rob Carnell <bertcarnell@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2024-01-21 17:30:02 UTC |
Generic Importance Plot
Description
Generic Importance Plot
Usage
importance(model_final, ...)
Arguments
model_final |
a model object |
... |
arguments passed to other methods |
Value
an object of type importance_plot
type |
the type of importance plot |
data |
the importance data required for the plot |
See Also
importance.glm importance.lm importance.cv.glmnet importance.survreg
Plot Variable Importance for a GLMNET model
Description
Plot Variable Importance for a GLMNET model
Usage
## S3 method for class 'cv.glmnet'
importance(model_final, model_data, form, dict = NA, nperm = 500, ...)
Arguments
model_final |
a model object |
model_data |
the data used to fit the model |
form |
the model formula |
dict |
a variable dictionary for plotting |
nperm |
the number of permutations used to calculate the importance |
... |
arguments passed to other methods |
Value
an object of type importance_plot
type |
the type of importance plot |
data |
the importance data required for the plot |
See Also
Examples
if (requireNamespace("glmnet", quietly = TRUE))
{
form <- formula(mpg ~ cyl*wt*hp)
mf <- model.frame(form, data = mtcars)
mm <- model.matrix(mf, mf)
gtest <- glmnet::cv.glmnet(x = mm, y = mtcars$mpg, family = "gaussian")
imp <- importance(gtest, mtcars, form, nperm = 50)
plot(imp)
}
GLM variable importance plot
Description
GLM variable importance plot
Usage
## S3 method for class 'glm'
importance(model_final, model_null, dict = NA, ...)
Arguments
model_final |
a model object |
model_null |
a glm object for the null model |
dict |
a dictionary to translate the model variables to plotting variables |
... |
arguments passed to other methods |
Value
an object of type importance_plot
type |
the type of importance plot |
data |
the importance data required for the plot |
See Also
Examples
gtest <- glm(mpg ~ cyl*wt*hp + gear + carb, data=mtcars, family=gaussian)
gtestreduced <- glm(mpg ~ 1, data=mtcars, family=gaussian)
imp <- importance(gtest, gtestreduced)
plot(imp)
gtest <- glm(mpg ~ cyl + wt + hp + gear + carb, data=mtcars, family=gaussian)
gtestreduced <- glm(mpg ~ 1, data=mtcars, family=gaussian)
imp <- importance(gtest, gtestreduced)
plot(imp)
gtest <- glm(vs ~ wt + disp + gear, data=mtcars, family=binomial(link="logit"))
gtestreduced <- glm(vs ~ 1, data=mtcars, family=binomial(link="logit"))
imp <- importance(gtest, gtestreduced)
plot(imp)
Linear Model variable importance plot
Description
Linear Model variable importance plot
Usage
## S3 method for class 'lm'
importance(model_final, model_null, dict = NA, ...)
Arguments
model_final |
a model object |
model_null |
a |
dict |
a dictionary to translate the model variables to plotting variables |
... |
arguments passed to other methods |
Value
an object of type importance_plot
type |
the type of importance plot |
data |
the importance data required for the plot |
See Also
Examples
gtest <- lm(mpg ~ cyl*wt*hp + gear + carb, data=mtcars)
gtestreduced <- lm(mpg ~ 1, data=mtcars)
imp <- importance(gtest, gtestreduced)
plot(imp)
gtest <- lm(mpg ~ cyl + wt + hp + gear + carb, data=mtcars)
gtestreduced <- lm(mpg ~ 1, data=mtcars)
imp <- importance(gtest, gtestreduced)
plot(imp)
Create a variable importance plot for a survreg model
Description
Create a variable importance plot for a survreg model
Usage
## S3 method for class 'survreg'
importance(model_final, model_data, dict = NA, nperm = 500, ...)
Arguments
model_final |
a model object |
model_data |
the data used to fit the model |
dict |
a plotting dictionary for models terms |
nperm |
the number of permutations used to calculate the importance |
... |
arguments passed to other methods |
Value
an object of type importance_plot
type |
the type of importance plot |
data |
the importance data required for the plot |
See Also
Examples
model_final <- survival::survreg(survival::Surv(futime, fustat) ~ ecog.ps*rx + age,
data = survival::ovarian,
dist = "weibull")
imp <- importance(model_final, survival::ovarian, nperm = 50)
plot(imp)
Importance Plot for the caret::train objects
Description
Importance Plot for the caret::train objects
Usage
## S3 method for class 'train'
importance(model_final, ...)
Arguments
model_final |
a model object |
... |
arguments passed to other methods |
Value
an object of type importance_plot
type |
the type of importance plot |
data |
the importance data required for the plot |
See Also
Examples
if (requireNamespace("caret", quietly = TRUE) &
requireNamespace("randomForest", quietly = TRUE))
{
model_final <- caret::train(x = subset(mtcars, select = -mpg), y = mtcars$mpg, method = "rf")
imp <- importance(model_final)
plot(imp)
}
Plot an Importance Plot object
Description
Plot an Importance Plot object
Usage
## S3 method for class 'importance_plot'
plot(
x,
plot = TRUE,
nvar = NA,
col_imp_alone = "#69BE28",
col_imp_cumulative = "#427730",
geom_bar_control = list(fill = "#69BE28"),
...
)
Arguments
x |
a |
plot |
boolean to determine if the plot is displayed, or just returned |
nvar |
the number of variables to plot in order of importance |
col_imp_alone |
the color used for the variance explained by each variable alone |
col_imp_cumulative |
the color used for the cumulative variance explained |
geom_bar_control |
list of arguments to control the plotting of |
... |
future arguments |
Value
the plot
Examples
gtest <- lm(mpg ~ cyl + wt + hp + gear + carb, data = mtcars)
gtestreduced <- lm(mpg ~ 1, data = mtcars)
imp <- importance(gtest, gtestreduced)
plot(imp)
gtest <- survival::survreg(survival::Surv(futime, fustat) ~ ecog.ps*rx + age,
data = survival::ovarian,
dist = "weibull")
imp <- importance(gtest, survival::ovarian, nperm = 50)
plot(imp)
Plot a Tornado Plot object
Description
Plot a Tornado Plot object
Usage
## S3 method for class 'tornado_plot'
plot(
x,
plot = TRUE,
nvar = NA,
xlabel = "Model Response",
sensitivity_colors = c("grey", "#69BE28"),
geom_bar_control = list(width = NULL),
geom_point_control = list(fill = "black", col = "black"),
...
)
Arguments
x |
a |
plot |
boolean to determine if the plot is displayed, or just returned |
nvar |
the number of variables to plot |
xlabel |
a label for the x-axis |
sensitivity_colors |
a two element character vector of the bar colors for a lower value and upper value |
geom_bar_control |
a list of |
geom_point_control |
a list of |
... |
future arguments |
Value
the plot
Examples
gtest <- lm(mpg ~ cyl*wt*hp, data = mtcars)
tp <- tornado(gtest, type = "PercentChange", alpha = 0.10, xlabel = "MPG")
plot(tp)
print data in an importance_plot
Description
print data in an importance_plot
Usage
## S3 method for class 'importance_plot'
print(x, ...)
Arguments
x |
the object to be printed |
... |
further arguments passed to |
Examples
gtest <- glm(vs ~ wt + disp + gear, data=mtcars, family=binomial(link="logit"))
gtestreduced <- glm(vs ~ 1, data=mtcars, family=binomial(link="logit"))
g <- importance(gtest, gtestreduced)
print(g)
print data in a tornado_plot
Description
print data in a tornado_plot
Usage
## S3 method for class 'tornado_plot'
print(x, ...)
Arguments
x |
the object to be printed |
... |
further arguments passed to |
Examples
gtest <- lm(mpg ~ cyl*wt*hp, data = mtcars)
tp <- tornado(gtest, type = "PercentChange", alpha = 0.10, xlabel = "MPG")
print(tp)
Quantile for Ordered Factors
Description
Quantile for Ordered Factors
Usage
## S3 method for class 'ordered'
quantile(x, probs = seq(0, 1, 0.25), ...)
Arguments
x |
an ordered factor |
probs |
the desired quatiles |
... |
arugments passed on |
Value
ordered factor levels at the desired quantiles
Examples
quantile(ordered(rep(c("C","B","A"), each=30), levels=c("C","B","A")),
probs <- seq(0, 1, 0.25))
Generic tornado plotting method
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
tornado(model, type, alpha, dict, ...)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
See Also
tornado.lm, tornado.glm, tornado.cv.glmnet, tornado.survreg, tornado.coxph, tornado.train
Cox Proportional Hazards Tornado Diagram
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
## S3 method for class 'coxph'
tornado(model, type = "PercentChange", alpha = 0.1, dict = NA, modeldata, ...)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
modeldata |
the data used to fit the model |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
Examples
gtest <- survival::coxph(survival::Surv(stop, event) ~ rx + size + number,
survival::bladder)
torn <- tornado(gtest, modeldata = survival::bladder, type = "PercentChange",
alpha = 0.10)
plot(torn, xlabel = "Risk")
GLMNET Tornado Diagram
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
## S3 method for class 'cv.glmnet'
tornado(
model,
type = "PercentChange",
alpha = 0.1,
dict = NA,
modeldata,
form,
s = "lambda.1se",
...
)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
modeldata |
the raw data used to fit the glmnet model |
form |
the model formula |
s |
Value(s) of the penalty parameter |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
See Also
Examples
if (requireNamespace("glmnet", quietly = TRUE))
{
form <- formula(mpg ~ cyl*wt*hp)
mf <- model.frame(form, data = mtcars)
mm <- model.matrix(form, data = mf)
gtest <- glmnet::cv.glmnet(x = mm, y= mtcars$mpg, family = "gaussian")
torn <- tornado(gtest, modeldata = mtcars, form = formula(mpg ~ cyl*wt*hp), s = "lambda.1se",
type = "PercentChange", alpha = 0.10)
plot(torn, xlabel = "MPG")
}
GLM Tornado Diagram
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
## S3 method for class 'glm'
tornado(model, type = "PercentChange", alpha = 0.1, dict = NA, ...)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
See Also
Examples
gtest <- glm(mpg ~ cyl*wt*hp, data = mtcars, family = gaussian)
torn <- tornado(gtest, type = "PercentChange", alpha = 0.10)
plot(torn, xlabel = "MPG")
Linear Model Tornado Diagram
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
## S3 method for class 'lm'
tornado(model, type = "PercentChange", alpha = 0.1, dict = NA, ...)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
See Also
Examples
gtest <- lm(mpg ~ cyl*wt*hp, data = mtcars)
torn <- tornado(gtest, type = "PercentChange", alpha = 0.10)
plot(torn, xlabel = "MPG")
Survreg Tornado Diagram
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
## S3 method for class 'survreg'
tornado(model, type = "PercentChange", alpha = 0.1, dict = NA, modeldata, ...)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
modeldata |
the data used to fit the model |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
See Also
Examples
gtest <- survival::survreg(survival::Surv(futime, fustat) ~ ecog.ps + rx,
survival::ovarian,
dist='weibull', scale=1)
torn <- tornado(gtest, modeldata = survival::ovarian, type = "PercentChange",
alpha = 0.10, xlabel = "futime")
plot(torn, xlabel = "Survival Time")
Caret Tornado Diagram
Description
A tornado plot is a visualization of the range of outputs expected from a variety of inputs, or alternatively, the sensitivity of the output to the range of inputs. The center of the tornado is plotted at the response expected from the mean of each input variable. For a given variable, the width of the tornado is determined by the range of the variable, a multiplicative factor of the variable, or a quantile of the variable. Variables are ordered vertically with the widest bar at the top and narrowest at the bottom. Only one variable is moved from its mean value at a time. Factors or categorical variables have also been added to these plots by plotting dots at the resulting output as each factor is varied through all of its levels. The base factor level is chosen as the input variable for the center of the tornado.
Usage
## S3 method for class 'train'
tornado(
model,
type = "PercentChange",
alpha = 0.1,
dict = NA,
class_number = NA,
...
)
Arguments
model |
a model object |
type |
|
alpha |
the level of change, the percentile level, or the number of standard deviations |
dict |
a dictionary to translate variables for the plot. The dictionary
must be a list or data.frame with elements |
class_number |
for classification models, which number of the class that will be plotted |
... |
further arguments, not used |
Value
a tornado_plot object
type |
the type of tornado plot |
data |
the data required for the plot |
family |
the model family if available |
See Also
Examples
if (requireNamespace("caret", quietly = TRUE) &
requireNamespace("randomForest", quietly = TRUE))
{
gtest <- caret::train(x = subset(mtcars, select = -mpg), y = mtcars$mpg, method = "rf")
torn <- tornado(gtest, type = "PercentChange", alpha = 0.10)
plot(torn, xlabel = "MPG")
}