vignettes/introduction_en.Rmd
introduction_en.Rmdexplor is an R package to allow interactive exploration
of multivariate analysis results.
For now on, the following analyses are supported :
| Analysis | Function | Package | Notes |
|---|---|---|---|
| Principal component analysis | PCA | FactoMineR | - |
| Correspondance analysis | CA | FactoMineR | - |
| Multiple correspondence analysis | MCA | FactoMineR | - |
| Principal component analysis | dudi.pca | ade4 | Qualitative supplementary variables are ignored |
| Correspondance analysis | dudi.coa | ade4 | - |
| Multiple correspondence analysis | dudi.acm | ade4 | Quantitative supplementary variables are ignored |
| Specific Multiple Correspondance Analysis | speMCA | GDAtools | - |
| Multiple Correspondance Analysis | mca | MASS | Quantitative supplementary variables are not supported |
| Principal Component Analysis | princomp | stats | Supplementary variables are ignored |
| Principal Component Analysis | prcomp | stats | Supplementary variables are ignored |
| Correspondance Analysis | textmodel_ca | quanteda.textmodels | Only coordinates are available |
The philosophy behind explor is to only be an
exploration interface which doesn’t really do anything by itself :
analysis and computations are made in your R script, and
explor only helps you visualizing the results. As such it
can not disrupt code execution and reproducibility.
For each type of analysis, explor launches a
shiny interactive Web interface which is displayed inside
RStudio or in your system Web browser. This interface provides a series
of tabs with interactive data and graphics.
These data and graphics are displayed with several “interactive”
features. Numerical results are shown as dynamic tables which are
sortable and searchable thanks to the DT package. Most
graphics are generated with the scatterD3 package which
provides the following features :
Usage is very simple : you just apply the explor()
function to the result object of one of the supported analysis
functions.
prcomp, princomp and
MASS::mca
To visualize and explore these functions results, just pass the
result object to explor().
Here is an example for a sample PCA with princomp :
explor supports the visualization of supplementary
individuals whose scores have been computed with predict.
You just have to add them as a supi element to your result
object.
Here is an example with prcomp :
pca <- prcomp(USArrests[6:50,], scale. = TRUE)
pca$supi <- predict(pca, USArrests[1:5,])
explor(pca)For MASS::mca, explor() also supports
qualitative supplementary variables. You must include their predicted
coordinates to a supv element. It’s also best to manually
add row names to the supidata, if any :
library(MASS)
mca <- MASS::mca(farms[4:20, 2:4], nf = 11)
supi_df <- farms[1:3, 2:4]
supi <- predict(mca, supi_df, type="row")
rownames(supi) <- rownames(supi_df)
mca$supi <- supi
mca$supv <- predict(mca, farms[4:20, 1, drop=FALSE], type="factor")
explor(mca)Note that the results of these three functions are quite limited : they provide variables and individuals coordinates, but no contributions or squared cosinus.
FactoMineR functions
Supported FactoMineR functions should work “out of the
box”. Just pass the result object to explor().
Example with a principal correspondence analysis from
FactoMineR::PCA :
library(FactoMineR)
data(decathlon)
pca <- PCA(decathlon[,1:12], quanti.sup = 11:12)
explor(pca)Example with a simple correspondence analysis from
FactoMiner::CA :
Example with a multiple correspondence analysis from
FactoMineR::MCA :
ade4 functions
ade4 functions should also work by directly passing the
object result to explor().
For example, to visualize a simple PCA results :
library(ade4)
data(deug)
pca <- dudi.pca(deug$tab, scale = TRUE, scannf = FALSE, nf = 5)
explor(pca)There’s a bit more work to be done if you want to display
supplementary elements, as ade4 don’t include them directly
in the results analysis.
For a principal component analysis, you have to compute supplementary
individuals (resp. variables) results with suprow (resp.
supcol) and add them manually as a supi (resp.
supv) element of your result object.
Here is an example of how to do this :
data(deug)
d <- deug$tab
sup_var <- d[-(1:10), 8:9]
sup_ind <- d[1:10, -(8:9)]
pca <- dudi.pca(d[-(1:10), -(8:9)], scale = TRUE, scannf = FALSE, nf = 5)
## Supplementary individuals
pca$supi <- suprow(pca, sup_ind)
## Supplementary variables
pca$supv <- supcol(pca, dudi.pca(sup_var, scale = TRUE, scannf = FALSE)$tab)
explor(pca)You have to do the same thing for supplementary elements in a multiple correspondence analysis :
data(banque)
d <- banque[-(1:100),-(19:21)]
ind_sup <- banque[1:100, -(19:21)]
var_sup <- banque[-(1:100),19:21]
acm <- dudi.acm(d, scannf = FALSE, nf = 5)
## Supplementary variables
acm$supv <- supcol(acm, dudi.acm(var_sup, scannf = FALSE, nf = 5)$tab)
## Supplementary individuals
acm$supi <- suprow(acm, ind_sup)
explor(acm)For simple correspondence analysis, you can add supplementary rows or
columns by adding their coordinates to supr and
supc elements of your result object :
GDAtools functions
GDAtools functions should also work by directly passing
the object result to explor().
For example, to visualize a speMCA results :
To display supplementary individuals, you have to compute their data
with the indsup function, and add them manually as a
supi element of your result object :
mca <- speMCA(Music[3:nrow(Music), 1:5], excl = c(3, 6, 9, 12, 15))
mca$supi <- indsup(mca, Music[1:2, 1:5])
explor(mca)To display supplementary variables, you have to compute their data
with the speMCA_varsup function and add them manually as a
supv element of your result object :
explor provides two different ways to export the
displayed plots.
To save the displayed plot as an SVG file, click on the Export to SVG button in the bottom of the left sidebar, or choose Export to SVG in the gear menu.
SVG is a vector graphics format, editable with softwares like Inkscape.
This SVG export may cause issues when used inside RStudio. As a
workaround, you can open explor in a browser (with Open
in Browser icon) before exporting.
Another way is to get the R code which allows to generate the current plot. This code can then be used in a script or a Rmarkdown document.
To do this, click on the Get R code button on the bottom of the left sidebar. A modal dialog should show up with the R code that you can then copy/paste.
Please note that this R code keeps track of the current plot zooming,
but not of any custom label positioning. If you want to keep those, you
have to first save them in a CSV file with Export labels
positions gear menu entry. Then, in your R script, read this file
in an object with read.csv and pass this object to the
export_labels_positions argument in the generated code
:
labels <- read.csv("position_labels.csv")
res <- explor::prepare_results(mca)
explor::MCA_var_plot(res, xax = 1, yax = 2,
var_sup = TRUE, , var_lab_min_contrib = 0,
col_var = "Variable", symbol_var = "Type",
size_var = NULL, size_range = c(10, 300),
labels_size = 10, point_size = 56,
transitions = TRUE, labels_positions = labels)explor is quite a young package, so there certainly are
bugs or problems. Thanks for reporting them by mail or by opening an issue on GitHub