condvis
Conditional Visualization for Statistical Models
Exploring fitted models by interactively taking 2-D and 3-D sections in data space.
Interactively exploring fitted models
Interactively take 2-D and 3-D sections in data space, showing where fitted
models intersect the section, and observed data near the section according to
a distance measure. See package website
for examples.
Works on Windows, Mac OS and Linux.
Requirements:
- Windows: the standard graphics device is sufficient.
- Mac OS: XQuartz device, website
- Linux: X11, included in some distributions.
Installation:
install.packages("condvis")Example to get started:
library(condvis)data(mtcars)m <- lm(mpg ~ wt + hp, data = mtcars)ceplot(data = mtcars, model = m, sectionvars = "hp")News
condvis: Conditional Visualisation for Statistical Models
News
Version 0.5-1 2018-09-13
- Changed check for zero variance (introduced in 0.5-0) to work for factors.
- Fixed S3 method treatment for
interpolate.
Version 0.5-0 2018-09-01
- Fixed bug where
arrangeCwould return pairs of the same variable if there were no bivariate relationships detected bysavingby2d. - Added a generic function
interpolate, with methods for standard data types. - Added an error for negative threshold dissimilarities in
similarityweight. - Fixed bug where variables with zero variance caused an error in
similarityweight.
Version 0.4-3 2017-10-19
- Fixed bug where
ceplot.shinydroppedxc.condto a vector and threw an error.
Version 0.4-2 2017-10-11
- New release for JSS publication. Small changes to documentation, and some default colours have been changed.
Version 0.4-0 2016-10-17
- The 2-D histogram in
plotxchas been made optional, and also allows a cap on the bin counts, to allow for skewed data. - In
similarityweight, the weight values are no longer discretised. All values between 0 and 1 are allowed. - New unexported function
weightcolorused to weight colours using similarity weight. - Fixed bug; not passing
pchto snapshot forceplot.interactive.
Version 0.3-5 2016-09-26
- In
plotxc, can chop off long tails for scatterplot/2-d histogram, controlled bytrimparameter. - pass plotting parameters for models from top-level
ceplotcall. See parametermodelpar. - allow
xlimandylimto be passed to section plot produced byplotxs. - corrected typo in
?similarityweight, sigma --> threshold. - added snapshot feature to
condtour. - changed treatment of conditioning predictors in
condtourto be more likeceplot, plus a hacky fix toupdate.xcplotforcondtourproviding predictor values instead of mouseclicks. - corrected problem in
condtour, where interpolated path omitted the final row. Also removed rownames from cluster centres. - diagnostic plot for
condtour,plotapnow shows approximate proportion of data visible, rather than approximate number of observations visible.
Version 0.3-4 2016-08-04
- fixed incorrect parameter names in demo files.
- fix to
plotxsto allowmainandmarto be specified.
Version 0.3-3 2016-07-04
- fixed treatment of
pchby condtour, was previously defaulting to 1 when it should have been 21 for background colours. - can now adjust
thresholdwhile usingcondtour. - added help link for
rsconnectpackage in Shiny application. - added fix to
update.xsplotwhere the fitted model was not updating properly, due to the section information remaining static.
Version 0.3-2
- fixed bug with
repforcolandpchwherelength.outwas omitted, causing huge delays forplotxs.
Version 0.3-1
- removed test which timed
similarityweight(for CRAN)
Version 0.3-0 2016-07-01
- removed daisy distance measure, and relevant contributors.
- renamed
visualweighttosimilarityweightand changed parameter names. - renamed parameters in
ceplot. - Shiny implementation completely rewritten, see R/ceplot.shiny.R.
- added Roxygen comments for documentation of exported functions.
- added update methods to parallel coordinates and scatterplot matrix condition selector plots.
condtourcan now handle mixed data (categorical and continuous).
Version 0.2-2 2016-05-01
- updated contributors for copied code.
- added simple example to README.
Version 0.2-1 23 Feb 2016
- initial release on CRAN.