[R] adding multiple fitted curves to xyplot graph

[Deepayan Sarkar]

On 8/16/06, GOUACHE David <D.GOUACHE at arvalisinstitutduvegetal.fr> wrote:
> Hello RHelpers,
>
> This may already have been answered, but despite days of scouring through the archives I haven't found it.
> My goal is to add multiple fitted curves to a plot.
> An example data set (a data frame named df in following code) is:
>
>              x1               y1     factor1
> 4       1298.25       0.00000000           1
> 5       1393.25       0.00000000           1
> 6       1471.50       0.04597701           1
> 7       1586.70       2.56908046           1
> 8       1692.10      11.14080460           1
> 9       1832.55      45.50459770           1
> 10      1928.30      65.56000000           1
> 11      2092.40     100.00000000           1
> 31      1202.90       0.00000000           2
> 41      1298.25       0.00000000           2
> 51      1393.25       0.37885057           2
> 61      1471.50       0.76839080           2
> 71      1586.70       7.75206897           2
> 81      1692.10      50.19448276           2
> 91      1832.55      94.08045977           2
> 101     1928.30     100.00000000           2
> 111     2092.40     100.00000000           2
> 14      1028.50       0.11111111           3
> 22      1106.40       0.04938272           3
> 32      1202.90       0.03448276           3
> 42      1298.25       0.34482759           3
> 52      1393.25       1.43850575           3
> 62      1471.50       1.96850575           3
> 72      1586.70      36.80597701           3
> 82      1692.10      92.83390805           3
> 92      1832.55     100.00000000           3
> 15      1028.50       0.09638554           4
> 23      1106.40       0.39988506           4
> 33      1202.90       0.49321839           4
> 43      1298.25       1.66045977           4
> 53      1393.25       7.51137931           4
> 63      1471.50      42.02724138           4
> 73      1586.70      99.12068966           4
> 83      1692.10     100.00000000           4
>
> I plot this with xyplot:
>
> trellis.par.set("background","white")

This doesn't do what you probably think it does.

> trellis.par.set(list(superpose.symbol=list(pch=c(15:17,21,25))))
> xyplot(y1 ~ x1, data=df, groups=factor1,
>             type = "p",
>             auto.key =
>             list(space = "right", points = TRUE, lines = FALSE))
>
> For each level of factor1 I fit a growth curve:
>
> fit.curve<-function(tab)
> {
>         res.fit<-nls(y1 ~ 100/(1+exp(((-log(81))/a)*(x1-b))), start=list(a=min(tab$x1[tab$y1>76],na.rm=T)-max(tab$x1[tab$y1<15],na.rm=T),b=tab$x1[abs(tab$y1-50)==min(abs(tab$y1-50),na.rm=T)][!is.na(tab$x1[abs(tab$y1-50)==min(abs(tab$y1-50),na.rm=T)])]),data=tab)
>         coef(res.fit)
> }
> by(df,list(df$factor1),fit.curve)
>
> I would like to add the 4 curves corresponding to these 4 fits to my graphic.
> The elegant way would be a custom panel function I suppose, but I haven't been able to write one up...
> Could someone help me out on this please?

Here's one solution:

xyplot(y1 ~ x1, data=df, groups=factor1,
       type = "p",
       panel = panel.superpose,
       panel.groups = function(x, y, ...) {
           panel.xyplot(x, y, ...)
           fm <-
               nls(y ~ 100 / (1 + exp(((-log(81)) / a) * (x - b))),
                   start =
                   list(a = min(x[y > 76], na.rm = TRUE) -
                        max(x[y < 15], na.rm = TRUE),
                        b = x[abs(y-50) == min(abs(y - 50),
                                 na.rm = TRUE)][
                                 !is.na(x[abs(y - 50) == min(abs(y-50),
                                             na.rm = TRUE)])]))
           fit.fun <- function(x) predict(fm, newdata = list(x = x))
           panel.curve(fit.fun, ...)
       },
       auto.key =
       list(space = "right", points = TRUE, lines = FALSE))


-Deepayan