[Implemented] Multi-pannel plots

[PavelBal]

It should be possible to combine multiple plots in one multi-pannel figure:

I was trying to figure out how the API should look like using the currently proposed functional approach (plotIndividualTimeProfile(dataCombined, plotConfiguration) and I cannot come up with any idea.

Here how it is implemented in esqlabsR:

Simulated and observed data are grouped in an object of the class DataMapping. As opposed to DataCombined, DataMapping does not only combines data and defines groupings, but also implements a plotting method and defines some plot properties:

• x-axis limits (can be automatically calculated)
• y-axis limits (can be automatically calculated)
• x-axis lable (default to Dimension [unit])
• y-axis lable (default to Dimension [unit])
• x/y log
• Title
• Show legend TRUE/FALSE
• legend position
• x units
• y units
• plotType - individual profile, population profile, simVsObs
• populationQuantiles for population profile
• For the data sets:
  • Line types (can be automatically calculated)
  • symbol types (can be automatically calculated)
  • colors (can be automatically calculated)

So basically DataMapping is a combination of DataCombined with a plot configuration. Calling DataMapping$plot() creates a figure as defined by the plotType.

To create a multipanel figure, multiple DataMapping objects are passed to the function plotMultiPanel

plotMultiPanel(c(dataMapping_timeProfile, dataMapping_simVsObs), plotConfiguration)

In this example, dataMapping_timeProfile has the plotType IndividualProfile, and dataMapping_simVsObs has the plotType simulated vs observed. plotConfiguration is an object that defines the output properties:

• output device (default device, png, or pdf)
• output file name
• output folder
• widht/height
• resolution
• point size
• nr of colums in a multi-pannel (calculated automatically if not specified)

Will be happy to hear your thoughts on how to implement it in ospsuite-r. @IndrajeetPatil @Yuri05 @msevestre

[Yuri05]

I don't understand this spec at all. Why don't we just say:

• we create N plots, each of them with its own configuration
• and then just combine them into a plot grid in a "standard" way, e.g. https://cran.r-project.org/web/packages/egg/vignettes/Ecosystem.html
  ?

So we could combine any types of plots (Individual time profiles, populatio time profiles, sensitivity plots, Gof plots, ddi ratio plots, custom plots, ...) in any order / layout.

[PavelBal]

I was not aware of the grid.arrange, so I think we are good. Might want to think if we want an additional wrapper around it, but it's low prio now.

More and more I get a feeling I was re-implementing ggplot xD

[PavelBal]

OK, how do I create the following plot?

I have three sub-plots, and want to arrange them. I also need to add title (SteadyState), and panel numbering (a, b, c)

[PavelBal]

@IndrajeetPatil

[IndrajeetPatil]

library(tlf)

# create a list of plots
ls_plots <- list(
  # first plot
  plotBoxWhisker(mtcars,
                 dataMapping = BoxWhiskerDataMapping$new(x = "am", y = "wt"), outliers = FALSE
  ),
  # second plot
  plotBoxWhisker(ToothGrowth,
                 dataMapping = BoxWhiskerDataMapping$new(x = "supp", y = "len")
  ),
  # third plot
  plotHistogram(x = mtcars$wt)
)

# create a new instance of plot config class
plotGridObj <- PlotGridConfiguration$new(ls_plots)

plotGridObj$title <- "SteadyState"
plotGridObj$nRows <- 2L
plotGridObj$tagLevels <- "a"

plotGrid(plotGridObj)

^{Created on 2022-05-12 by the reprex package (v2.0.1.9000)}

[PavelBal]

This does not work with plots created with plotIndividualTimeProfile, so it has to be implemented.

[IndrajeetPatil]

Works just fine for me.

Are you sure you are using the development version of tlf with the latest commit?

library(ospsuite)
library(ospsuite.utils)
library(tlf)

# import observed data (will return a list of DataSet objects)
dataSet <- loadDataSetsFromExcel(
  xlsFilePath = "CompiledDataSet.xlsx",
  importerConfiguration = loadDataImporterConfiguration("ImporterConfiguration.xml")
)

# simulation results
sim <- importResultsFromCSV(
  simulation = loadSimulation("Stevens_2012_placebo_indiv_results.pkml"),
  filePaths = "Stevens_2012_placebo_indiv_results.csv"
)

# create a `DataCombined` object and add observed and simulated data
myDataComb <- DataCombined$new()
myDataComb$addDataSets(dataSet)
myDataComb$addSimulationResults(sim,
  quantitiesOrPaths = list(
    "Organism|Lumen|Stomach|Metformin|Gastric retention",
    "Organism|Lumen|Stomach|Metformin|Gastric retention distal",
    "Organism|Lumen|Stomach|Metformin|Gastric retention proximal"
  )
)

# set groupings
myDataComb$setGroups(
  list(
    "Stevens_2012_placebo.Placebo_distal",
    "Organism|Lumen|Stomach|Metformin|Gastric retention distal",
    "Stevens_2012_placebo.Placebo_proximal",
    "Organism|Lumen|Stomach|Metformin|Gastric retention proximal",
    "Stevens_2012_placebo.Placebo_total",
    "Organism|Lumen|Stomach|Metformin|Gastric retention"
  ),
  list(
    # group 1
    "Stevens 2012 solid distal",
    "Stevens 2012 solid distal",
    # group 2
    "Stevens 2012 solid proximal",
    "Stevens 2012 solid proximal",
    # group 3
    "Stevens 2012 solid total",
    "Stevens 2012 solid total"
  )
)

myPC <- DefaultPlotConfiguration$new()
myPC$yUnit <- "%"

p1 <- plotIndividualTimeProfile(myDataComb, myPC)

myPC$linesLinetype <- names(tlf::Linetypes)
p2 <- plotIndividualTimeProfile(myDataComb, myPC)

myPC$linesLinetype <- tlf::Linetypes$dashed
myPC$pointsShape <- "circle"
p3 <- plotIndividualTimeProfile(myDataComb, myPC)

ls_plots <- list(p1, p2, p3)

# create a new instance of plot config class
plotGridObj <- PlotGridConfiguration$new(ls_plots)

plotGridObj$title <- "SteadyState"
plotGridObj$nRows <- 2L
plotGridObj$tagLevels <- "a"

plotGrid(plotGridObj)

[PavelBal]

Ok I hate R's lists.

How can I dynamically add plot objects to a list?

plots <- list()
plot <- plotIndividualTimeProfile(dataCombined = currentDataCombined, defaultPlotConfiguration = plotConfigurationScenario)

plots <- c(plots, plot)

And it does not work any more.

[IndrajeetPatil]

You can use append():

plots <- list()
plot <- "x"
plots <- append(plots, plot)
plots
#> [[1]]
#> [1] "x"

^{Created on 2022-05-12 by the reprex package (v2.0.1.9000)}

[PavelBal]

Nope,

>   plots <- list()
> plots <- append(plots, plot)
> length(plots)
[1] 11

Actually, append uses c, so no difference to expect.

[IndrajeetPatil]

Is dynamically updating the list not what you wanted? append() seems to do the trick:

plots <- list()
x <- letters[1:3]

for (elem in x) {
  plots <- append(plots, elem)
}

plots
#> [[1]]
#> [1] "a"
#> 
#> [[2]]
#> [1] "b"
#> 
#> [[3]]
#> [1] "c"

^{Created on 2022-05-12 by the reprex package (v2.0.1.9000)}

[PavelBal]

Just try it with your own example with the plots:

dataSet <- loadDataSetsFromExcel(
  xlsFilePath = "CompiledDataSet.xlsx",
  importerConfiguration = loadDataImporterConfiguration("ImporterConfiguration.xml")
)
# create a `DataCombined` object and add observed and simulated data
myDataComb <- DataCombined$new()
myDataComb$addDataSets(dataSet)
myPC <- DefaultPlotConfiguration$new()
myPC$yUnit <- "%"

p1 <- plotIndividualTimeProfile(myDataComb, myPC)

ls_plots <- list()
ls_plots <- append(ls_plots, p1)

plotGridObj <- PlotGridConfiguration$new(ls_plots)
plotGrid(plotGridObj)

> plotGrid(plotGridObj)
 Error: Only know how to add ggplots and/or grobs
3.
stop("Only know how to add ggplots and/or grobs", call. = FALSE)
2.
patchwork::wrap_plots(plotGridConfiguration$plotList, ncol = plotGridConfiguration$nColumns, 
nrow = plotGridConfiguration$nRows, byrow = plotGridConfiguration$byRow, 
widths = plotGridConfiguration$widths, heights = plotGridConfiguration$heights, 
guides = plotGridConfiguration$guides, design = plotGridConfiguration$design) at plot-grid.R#72
1.
plotGrid(plotGridObj)

so, no, it does not work.

[IndrajeetPatil]

That should fix it.

ls_plots <- list()
ls_plots <- append(ls_plots, list(p1))

[PavelBal]

This is so ugly. Can we implement a method addPlot for PlotGridConfiguration class?

[msevestre]

yes!!!
or not use list :)