How can I rewrite this code so that it uses plyr/ddply as intended?
Background
I have a dataframe of probability distributions that I would like to calculate statistical summaries for:
priors <- structure(l开发者_开发技巧ist(name = c("theta1", "theta2", "theta3", "theta4",
"theta5"), distn = c("gamma", "beta", "lnorm", "weibull", "gamma"),
parama = c(2.68, 4, 1.35, 1.7, 2.3), paramb = c(0.084, 7.2, 0.69, 0.66, 3.9),
another_col = structure(c(3L, 4L, 5L, 1L, 2L
), .Label = c("1", "2", "a", "b", "c"), class = "factor")),
.Names = c("name", "distn", "parama", "paramb", "another_col"), row.names = c("1",
"2", "3", "4", "5"), class = "data.frame")
Approach
Step 1: I wrote a function to calculate the summaries and returning mean(lcl, ucl)
summary.stats <- function(distn, A, B) {
if (distn == 'gamma' ) ans <- c(A*B, qgamma(c(0.05, 0.95), A[ ], B))
if (distn == 'lnorm' ) ans <- c(exp(A + 1/2 * B^2), qlnorm(c(0.05, 0.95), A, B))
if (distn == 'beta' ) ans <- c(A/(A+B), qbeta( c(0.05, 0.95), A, B))
if (distn == 'weibull') ans <- c(mean(rweibull(10000,A,B)), qweibull(c(0.05, 0.95), A, B))
if (distn == 'norm' ) ans <- c(A, qnorm( c(0.05, 0.95), A, B))
ans <- (signif(ans, 2))
return(paste(ans[1], ' (', ans[2], ', ', ans[3],')', sep = ''))
}
Step 2: I would like to add a new column to my dataframe called stats
priors$stats <- ddply(priors,
.(name, distn, parama, paramb),
function(x) summary.stats(x$distn, x$parama, x$paramb))$V1
Question 1:
what is the proper way to do this? I get an error when I try
ddply(priors,
.(name, distn, parama, paramb),
transform,
stats = function(x) summary.stats(x$distn, x$parama, x$paramb))
Question 2: (extra credit)
Is there a more efficient way to code the summary.stats
function, i.e., with less 'if's'?
update
Thanks to Shane and Joshua for clearing this up for me.
I also found a question that should be helpful for others trying to do a plyr operation on every row of a dataframe
Here's a cleaned-up version of your summary.stats
that uses switch
instead. I also added the name "stats" to the output, since that seems to be the thing tripping you up.
summaryStats <- function(distn, A, B) {
CI <- c(0.05, 0.95)
FUN <- get(paste("q",distn,sep=""))
ans <- switch(distn,
gamma = A*B,
lnorm = exp(A + 1/2 * B^2),
beta = A/(A+B),
weibull = mean(rweibull(10000,A,B)),
norm = A)
ans <- c(ans, FUN(CI, A, B))
ans <- (signif(ans, 2))
out <- c(stats=paste(ans[1], ' (', ans[2], ', ', ans[3],')', sep=''))
return(out)
}
I'm not sure how to do this with plyr
, but you can do it with boring ol' sapply
like this:
priors$stats <- sapply(1:nrow(priors),
function(i) with(priors[i,], summaryStats(distn, parama, paramb) ))
I could be missing something, but using Josh's function and your data, this works fine.
priors <- ddply(priors,
.(name, distn, parama, paramb),
function(x) summaryStats(x$distn, x$parama, x$paramb))
colnames(priors)[5] <- "stats"
What do you want your output to look like?
> priors
name distn parama paramb stats
1 theta1 gamma 2.68 0.084 0.23 (7.8, 69)
2 theta2 beta 4.00 7.200 0.36 (0.15, 0.6)
3 theta3 lnorm 1.35 0.690 4.9 (1.2, 12)
4 theta4 weibull 1.70 0.660 0.59 (0.12, 1.3)
5 theta5 gamma 2.30 3.900 9 (0.12, 1.3)
Edit
Sorry, didn't read your whole comment. Then this should work (in my example here, I leave out one column):
ddply(priors, .(distn, parama, paramb), function(x)
data.frame(x, stats=summaryStats(x$distn, x$parama, x$paramb)))
精彩评论