simplify source code

Hi:

Here's one way you could do it. I manufactured some fake data with a
simple model to illustrate. This assumes you are using the same model
formula with the same starting values and remaining arguments for each
response.

dg <- data.frame(x = 1:10, y1 = sort(abs(rnorm(10))),
                  y2 = sort(abs(rnorm(10))), y3 = sort(abs(rnorm(10))))

# Model: y = b0 + b1 exp(x/theta)
vars <- c('y1', 'y2', 'y3')

# Function to create the model formula by plugging in the
# response y and run the model
mfun <- function(y) {
     form <- as.formula(paste(y, 'cbind(1, exp(x/th))', sep = ' ~ '))
     nls(form, data = dg, start = list(th = 0.3), algorithm = 'plinear')
    }

# Generate a list of model objects:
mlist <- lapply(vars, mfun)

# To see what they contain:
str(mlist[[1]])
str(summary(mlist[[1]]))

# Extract a few features from each:
# The first two return matrices, the third returns a list

do.call(rbind, lapply(mlist, function(m) coef(m)))
do.call(rbind, lapply(mlist, function(m) deviance(m)))
lapply(mlist, function(m) summary(m)$cov.unscaled)

To get more control over the output format, the plyr package can come
in handy. For example, to get data frames for the first two
extractions above, one would do

library('plyr')
ldply(mlist, function(m) coef(m))
ldply(mlist, function(m) deviance(m))

# ldply() means list input, data frame output (ld).
# For the third extraction, one has a list input and a list output:
llply(mlist, function(m) summary(m)$cov.unscaled)

HTH,
Dennis

simplify source code

Thread (4 messages)