drc:: optim initial value in 'vmmin' is not finite

Question

I want to fit log-logistic regression for the following data in drc R package. However, my code throws the following error.

df1 <-
  structure(list(Temp = c(15L, 15L, 15L, 15L, 15L, 15L, 15L, 15L, 
      15L, 20L, 20L, 20L, 20L, 20L, 20L, 25L, 25L, 25L, 25L, 30L, 30L, 
      30L, 30L, 35L, 35L, 35L, 35L, 40L, 40L, 40L, 40L), Start = c(0L, 
      24L, 48L, 72L, 96L, 120L, 144L, 168L, 192L, 0L, 24L, 48L, 72L, 
      96L, 120L, 0L, 24L, 48L, 72L, 0L, 24L, 48L, 72L, 0L, 24L, 48L, 
      72L, 0L, 24L, 48L, 72L), End = c(24, 48, 72, 96, 120, 144, 168, 
      192, Inf, 24, 48, 72, 96, 120, Inf, 24, 48, 72, 96, 24, 48, 72, 
      Inf, 24, 48, 72, Inf, 24, 48, 72, Inf), Germinated = c(0L, 0L, 
      1L, 3L, 3L, 12L, 14L, 12L, 15L, 0L, 11L, 27L, 15L, 3L, 4L, 2L, 
      30L, 15L, 13L, 6L, 43L, 7L, 4L, 5L, 48L, 3L, 4L, 0L, 31L, 21L, 
      8L)), .Names = c("Temp", "Start", "End", "Germinated"), row.names = c(NA, 
      -31L), class = "data.frame")

library(drc)

fm1 <-
  drm(
        formula = Germinated ~ Start + End 
      , data    = df1
      , fct     = LL.2()
      , type    = "event" 
      , control   = drmc(
                            constr      = FALSE
                            , errorm      = TRUE
                            , maxIt       = 1500
                            , method      = "BFGS"
                            , noMessage   = FALSE
                            , relTol      = 1e-07
                            , rmNA        = FALSE
                            , useD        = FALSE
                            , trace       = FALSE
                            , otrace      = FALSE
                            , warnVal     = -1
                            , dscaleThres = 1e-15
                            , rscaleThres = 1e-15
                            )

          )

summary(fm1)

You absolutely right @G.Grothendieck. I already tried `curveid = factor(Temp)` and had the same error. Any thoughts. — MYaseen208, Jul 03 '16 at 12:29
This gives results (also try with `separate=TRUE` and `separate = FALSE` and compare to subsetting the data to a single temperature and just running that) but you do get separate coefficients for each temperature in all cases: `drm(Germinated ~ Start + End, data = df1, curveid = Temp, fct = LL.2(), type = "event")` — G. Grothendieck, Jul 03 '16 at 12:36

dww · Accepted Answer · 2016-07-08T14:38:46.963

You need to group the data by Temp, because the time periods are repeated for each value of Temp. curveid = Temp does the trick:

fm1 <-drm(data = df1, curveid = Temp,
          formula = Germinated ~ Start + End,  fct = LL.2(), type = "event", 
          control = drmc(constr = FALSE, errorm = TRUE, maxIt = 1500, method = "BFGS", 
                         noMessage = FALSE, relTol = 1e-07, rmNA = FALSE, useD = FALSE, 
                         trace = FALSE, otrace = FALSE, warnVal = -1, dscaleThres = 1e-15, rscaleThres = 1e-15))

summary(fm1)

Model fitted: Log-logistic (ED50 as parameter) with lower limit at 0 and upper limit at 1 (2 parms)

Parameter estimates:

      Estimate Std. Error   t-value p-value
b:15  -6.03055    0.78915  -7.64179       0
b:20  -4.96450    0.60740  -8.17338       0
b:25  -4.43973    0.54904  -8.08639       0
b:30  -4.80876    0.60792  -7.91025       0
b:35  -5.45991    0.69159  -7.89467       0
b:40  -5.43892    0.79772  -6.81811       0
e:15 162.33568    6.10473  26.59177       0
e:20  64.71588    3.08660  20.96674       0
e:25  48.23883    2.68278  17.98090       0
e:30  36.38415    2.04252  17.81337       0
e:35  35.07398    1.85537  18.90405       0
e:40  48.44494    2.21375  21.88366       0

Thanks @dww for useful answer. Would appreciate if you guide me how to fit `drm` model for the interaction of `Temp` and `Variety`, I tried `curveid = Temp:Variety` but throws error: `Error in Temp:Variety : NA/NaN argument`. Any thoughts, please. — MYaseen208, Jul 08 '16 at 09:23
I can certainly show you how to do this - but I think that this question is something that others may find useful in its own right. Therefore, can I suggest that you ask this as a new question. That way the solution becomes searchable to others who also want to know how to group on more than one column. If I answer it here, the solution will be rather hidden from other SO users. — dww, Jul 08 '16 at 14:24
Thanks @dww for your suggestion. New question is [here](http://stackoverflow.com/q/38269519/707145). — MYaseen208, Jul 08 '16 at 14:47

drc:: optim initial value in 'vmmin' is not finite

1 Answers1