[R] Overlaying lattice graphs (continued)
hadley wickham
h.wickham at gmail.com
Fri Jun 22 21:40:26 CEST 2007
Hi Sebastian,
I think the following does what you want:
library(ggplot2)
names(mydata) <- tolower(names(mydata))
obs <- rename(subset(mydata, model=="A", -predicted), c("observed" = "value"))
obs$model <- factor("observed")
pred <- rename(mydata[, -5], c("predicted" = "value"))
all <- rbind(obs, pred)
ggplot(all, aes(x = time, y = value, colour=model)) +
geom_point(data = subset(all, model != "Observed")) +
geom_line(data= subset(all, model == "Observed")) +
facet_grid(. ~ individuals)
Hadley
On 6/22/07, Sébastien <pomchip at free.fr> wrote:
> Hi Deepayan,
>
> The following code creates a dummy dataset which has the same similar as
> my usual datasets. I did not try to implement the changes proposed by
> Hadley, hoping that a solution can be found using the original dataset.
>
> ######### My code
>
> # Creating dataset
>
> nPts<-10 # number of time points
> nInd<-6 # number of individuals
> nModel<-3 # number of models
>
> TimePts<-rep(1:nPts,nInd*nModel) #
> creates the "Time" column
> Coef<-rep(rnorm(6,0.1,0.01),each=nPts,nModel) # Creates a
> vector of coefficients for generating the observations
> Obs<-10*exp(-Coef*TimePts) #
> creates the observations
>
> for (i in 1:60){
> Pred[i]<-jitter(10*exp(-Coef[i]*TimePts[i]))
> Pred[i+60]<-jitter(5)
> Pred[i+120]<-jitter(10-Coef[i+120]*TimePts[i])
> }
> # creates the predicted values
>
> colPlot<-rep(1,nPts*nInd*nModel)
> # creates the "Plot" column
> colModel<-gl(nModel,nPts*nInd,labels=c("A","B","C")) #
> creates the "Model" column
> colID<-gl(nInd,nPts,nPts*nInd*nModel)
> # creates the "ID" column
>
> mydata<-data.frame(colPlot,colModel,colID,TimePts,Obs,Pred)
> # creates the dataset
> names(mydata)<-c("Plot","Model","Individuals","Time","Observed","Predicted")
>
> # Plotting as indicated by Deepayan
>
>
> xyplot(Observed + Predicted ~ Time | Individuals + Model,
> data = mydata,
> panel = panel.superpose.2, type = c("p", "l"),
> layout = c(0, nlevels(mydata$Individuals))) #,
> #<...>)
>
> ####### End of code
>
> This codes is not exactly what I am looking for, although it is pretty
> close. In the present case, I would like to have a Trellis plot with 6
> panels (one for each individual), where the Observations and the
> Predicted are plotted as symbols and lines, respectively. All three
> models should be plotted on the same panel. Unfortunately, it looks to
> me as 3 successives xyplots are created by the code above but only the
> last one remains displayed. I tried to play with
> panel.superpose,panel.superpose.2 and type, without much success.
>
> I also tried the following code that creates 18 panels and distinguish
> all (Individuals,Model) couples... so, not what I want.
>
> xyplot(Observed + Predicted ~ Time | Individuals+Model, data = mydata,
> type = c("p", "l"), distribute.type = TRUE)
>
> Sebastien
>
>
> Deepayan Sarkar a écrit :
> > On 6/21/07, Sébastien <pomchip at free.fr> wrote:
> >> Hi Hadley,
> >>
> >> Hopefully, my dataset won't be too hard to changed. Can I modify the
> >> aspect of each group using your code (symbols for observed and lines for
> >> predicted)?
> >>
> >> Sebastien
> >>
> >> hadley wickham a écrit :
> >> > Hi Sebastian,
> >> >
> >> > I think you need to rearrange your data a bit. Firstly, you need to
> >> > put observed on the same footing as the different models, so you would
> >> > have a new column in your data called value (previously observed and
> >> > predicted) and a new model type ("observed"). Then you could do:
> >
> > Yes, and ?make.groups (and reshape of course) could help with that.
> > This might not be strictly necessary though.
> >
> > However, I'm finding your pseudo-code confusing. Could you create a
> > small example data set that can be used to try out some real code?
> > Just from your description, I would have suggested something like
> >
> > xyplot(Observed + Predicted ~ Time | Individuals + Model,
> > data = mydata,
> > panel = panel.superpose.2, type = c("p", "l"),
> > layout = c(0, nlevels(mydata$Individuals)),
> > <...>)
> >
> > If all you want is to plot one page at a time, there are easier ways
> > to do that.
> >
> > -Deepayan
> >
> >> >
> >> > xyplot(value ~ time | individauls, data=mydata, group=model)
> >> >
> >> > Hadley
> >> >
> >> >
> >> > On 6/21/07, Sébastien <pomchip at free.fr> wrote:
> >> >> Dear R Users,
> >> >>
> >> >> I recently posted an email on this list about the use of
> >> data.frame and
> >> >> overlaying multiple plots. Deepayan kindly indicated to me the
> >> >> panel.superposition command which worked perfectly in the context
> >> of the
> >> >> example I gave.
> >> >> I'd like to go a little bit further on this topic using a more
> >> complex
> >> >> dataset structure (actually the one I want to work on).
> >> >>
> >> >> >mydata
> >> >> Plot Model Individuals Time Observed
> >> >> Predicted
> >> >> 1 1 A 1 0.05
> >> >> 10 10.2
> >> >> 2 1 A 1 0.10
> >> >> 20 19.5
> >> >> etc...
> >> >> 10 1 B 1 0.05 10
> >> >> 9.8
> >> >> 11 1 B 1 0.10 20
> >> >> 20.2
> >> >> etc...
> >> >>
> >> >> There are p "levels" in mydata$Plot, m in mydata$Model, n in
> >> >> mydata$Individuals and t in mydata$Time (Note that I probably use the
> >> >> word levels improperly as all columns are not factors). Basically,
> >> this
> >> >> dataset summarizes the t measurements obtained in n individuals as
> >> well
> >> >> as the predicted values from m different modeling approaches
> >> (applied to
> >> >> all individuals). Therefore, the observations are repeated m times in
> >> >> the Observed columns, while the predictions appears only once for a
> >> >> given model an a given individual.
> >> >>
> >> >> What I want to write is a R batch file creating a Trellis graph,
> >> where
> >> >> each panel corresponds to one individual and contains the
> >> observations
> >> >> (as scatterplot) plus the predicted values for all models (as
> >> lines of
> >> >> different colors)... $Plot is just a token: it might be used to not
> >> >> overload graphs in case there are too many tested models. The fun
> >> part
> >> >> is that the values of p, m, n and t might vary from one dataset to
> >> the
> >> >> other, so everything has to be coded dynamically.
> >> >>
> >> >> For the plotting part I was thinking about having a loop in my code
> >> >> containing something like that:
> >> >>
> >> >> for (i in 1:nlevels(mydata$Model)) {
> >> >>
> >> >> subdata<-subset(mydata,mydata$Model=level(mydata$Model)[i])
> >> >> xyplot(subset(Observed + Predicted ~ Time | Individuals, data =
> >> >> subdata) #plus additionnal formatting code
> >> >>
> >> >> }
> >> >>
> >> >> Unfortunately, this code simply creates a new Trellis plot instead of
> >> >> adding the model one by one on the panels. Any idea or link to a
> >> useful
> >> >> command will wellcome.
> >> >>
> >> >> Sebastien
> >> >>
> >> >> ______________________________________________
> >> >> R-help at stat.math.ethz.ch mailing list
> >> >> https://stat.ethz.ch/mailman/listinfo/r-help
> >> >> PLEASE do read the posting guide
> >> >> http://www.R-project.org/posting-guide.html
> >> >> and provide commented, minimal, self-contained, reproducible code.
> >> >>
> >> >
> >> >
> >>
> >> ______________________________________________
> >> R-help at stat.math.ethz.ch mailing list
> >> https://stat.ethz.ch/mailman/listinfo/r-help
> >> PLEASE do read the posting guide
> >> http://www.R-project.org/posting-guide.html
> >> and provide commented, minimal, self-contained, reproducible code.
> >>
>
More information about the R-help
mailing list