Predict after running the mlogit function in R Here's what I want to do, but there seem to be no predict method for the mlogit.  Any ideas?
library(mlogit)
data("Fishing", package = "mlogit")
Fish <- mlogit.data(Fishing, varying = c(2:9), shape = "wide", choice = "mode")

Fish_fit<-Fish[-1,]
Fish_test<-Fish[1,]
m <- mlogit(mode ~price+ catch | income, data = Fish_fit)
predict(m,newdata=Fish_test)

 A: The mlogit package does have a predict() method, at least in the version I'm using ( 0.2-3 with R 2.15.3).
The code put up by @Zach has one error in it. The "long format" data used by mlogit() has one row for each alternative; this is the format created by the mlogit.data() function. Therefore to get a prediction for the first case you need to pull out all the rows for that case, and there are 4:
Fish_fit<-Fish[-(1:4),]
Fish_test<-Fish[1:4,]
m <- mlogit(mode ~price+ catch | income, data = Fish_fit)
predict(m,newdata=Fish_test)

which gives a good result. 
A: After quite a lot of effort in trying to use the predict function for the population, I think I can add a few insights to all your answers. 
The predict function of mlogit works fine, you just have to make some adjustments and be sure that the following things are taken care of:


*

*The newdata (as expected) should include exactly the same data as the sample used for the estimation of the model. This means that one should check for "hidden" properties of the data (such as a factor that inherits levels that do not exist -droplevel can be useful in this case-, or not introduced in the sample factors, or a wrong colname etc.). 

*You have to make an arbitrary choice in your newdata (if it does not exist) something that can be easily done using the sample function:
MrChoice <-sample(c("Car", "Bus", "Walk"),nrow(datase),replace=TRUE, prob = c(0.5, 0.4, 0.1))
mynewData$mode<-MrChoice


*The next required step is to again transform the data to mlogit data, using the same function as used for the sample data, for example:
ExpData3<- mlogit.data(mynewData, shape="wide", choice = "mode",sep=".",id = "TripID")


*The final step would be the actual prediction using the predict function. 
resulted<-predict(ml1,newdata=ExpData3)

A: To answer my own question, I've moved over to using the 'glmnet' package to fit my multinomial logits, which has the added advantage of using the lasso or elastic net to regularize my independent variables.  glmnet seems to be a much more 'finished' packaged than mlogit, complete with a 'predict' function.
A: Here's useful trick: Add the data you want to predict to your original estimation sample, but use the weights variable to set the weight of those new observations to zero. Estimate the model (with the new observations weighted to zero), and get the predictions from the "probabilities" output. That way you can bypass the predict function, which is a mess.
A: mlogit has a predict function, but I found it very difficult to use. I wrote my own very ugly set of functions for an implementation that I have. Anyone is welcome to use or improve them, stored on my github profile.
A: I'm pretty sure this is easily done with the given mlogit package by using the fitted function and then the standard R predict function.  As chl pointed out, although I haven't done it myself yet (at least not the predict), is exampled in the package vignettes here on pg 29.
