What you might need to do is create a character string with your formula in it (looping through pairs of variables and using paste or sprint) then convert that to a formula using the as.formula function.
-- Gregory (Greg) L. Snow Ph.D. Statistical Data Center Intermountain Healthcare greg.s...@imail.org 801.408.8111 > -----Original Message----- > From: Matthew Douglas [mailto:matt.dougla...@gmail.com] > Sent: Thursday, March 03, 2011 2:09 PM > To: Greg Snow > Cc: r-help@r-project.org > Subject: Re: [R] Regression with many independent variables > > Thanks greg, > > that formula was exactly what I was looking for. Except now when I > run it on my data I get the following error: > > "Error in model.matrix.default(mt, mf, contrasts) : cannot allocate > vector of length 2043479998" > > I know there are probably many 2-way interactions that are zero so I > thought I could save space by removing these. Is there some way that > can just delete all the two way interactions that are zero and keep > the columns that have non-zero entries? I think that will > significantly cut down the memory needed. Or is there just another way > to get around this? > > thanks, > Matt > > On Tue, Mar 1, 2011 at 3:56 PM, Greg Snow <greg.s...@imail.org> wrote: > > You can use ^2 to get all 2 way interactions and ^3 to get all 3 way > interactions, e.g.: > > > > lm(Sepal.Width ~ (. - Sepal.Length)^2, data=iris) > > > > The lm.fit function is what actually does the fitting, so you could > go directly there, but then you lose the benefits of using . and ^. > The Matrix package has ways of dealing with sparse matricies, but I > don't know if that would help here or not. > > > > You could also just create x'x and x'y matricies directly since the > variables are 0/1 then use solve. A lot depends on what you are doing > and what questions you are trying to answer. > > > > -- > > Gregory (Greg) L. Snow Ph.D. > > Statistical Data Center > > Intermountain Healthcare > > greg.s...@imail.org > > 801.408.8111 > > > > > >> -----Original Message----- > >> From: Matthew Douglas [mailto:matt.dougla...@gmail.com] > >> Sent: Tuesday, March 01, 2011 1:09 PM > >> To: Greg Snow > >> Cc: r-help@r-project.org > >> Subject: Re: [R] Regression with many independent variables > >> > >> Hi Greg, > >> > >> Thanks for the help, it works perfectly. To answer your question, > >> there are 339 independent variables but only 10 will be used at one > >> time . So at any given line of the data set there will be 10 non > zero > >> entries for the independent variables and the rest will be zeros. > >> > >> One more question: > >> > >> 1. I still want to find a way to look at the interactions of the > >> independent variables. > >> > >> the regression would look like this: > >> > >> y = b12*X1X2 + b23*X2X3 +...+ bk-1k*Xk-1Xk > >> > >> so I think the regression in R would look like this: > >> > >> lm(MARGIN, P235:P236+P236:P237+....,weights = Poss, data = adj0708), > >> > >> my problem is that since I have technically 339 independent > variables, > >> when I do this regression I would have 339 Choose 2 = approx 57000 > >> independent variables (a vast majority will be 0s though) so I dont > >> want to have to write all of these out. Is there a way to do this > >> quickly in R? > >> > >> Also just a curious question that I cant seem to find to online: > >> is there a more efficient model other than lm() that is better for > >> very sparse data sets like mine? > >> > >> Thanks, > >> Matt > >> > >> > >> On Mon, Feb 28, 2011 at 4:30 PM, Greg Snow <greg.s...@imail.org> > wrote: > >> > Don't put the name of the dataset in the formula, use the data > >> argument to lm to provide that. A single period (".") on the right > >> hand side of the formula will represent all the columns in the data > set > >> that are not on the left hand side (you can then use "-" to remove > any > >> other columns that you don't want included on the RHS). > >> > > >> > For example: > >> > > >> >> lm(Sepal.Width ~ . - Sepal.Length, data=iris) > >> > > >> > Call: > >> > lm(formula = Sepal.Width ~ . - Sepal.Length, data = iris) > >> > > >> > Coefficients: > >> > (Intercept) Petal.Length Petal.Width > >> Speciesversicolor > >> > 3.0485 0.1547 0.6234 > - > >> 1.7641 > >> > Speciesvirginica > >> > -2.1964 > >> > > >> > > >> > But, are you sure that a regression model with 339 predictors will > be > >> meaningful? > >> > > >> > -- > >> > Gregory (Greg) L. Snow Ph.D. > >> > Statistical Data Center > >> > Intermountain Healthcare > >> > greg.s...@imail.org > >> > 801.408.8111 > >> > > >> > > >> >> -----Original Message----- > >> >> From: r-help-boun...@r-project.org [mailto:r-help-bounces@r- > >> >> project.org] On Behalf Of Matthew Douglas > >> >> Sent: Monday, February 28, 2011 1:32 PM > >> >> To: r-help@r-project.org > >> >> Subject: [R] Regression with many independent variables > >> >> > >> >> Hi, > >> >> > >> >> I am trying use lm() on some data, the code works fine but I > would > >> >> like to use a more efficient way to do this. > >> >> > >> >> The data looks like this (the data is very sparse with a few 1s, > -1s > >> >> and the rest 0s): > >> >> > >> >> > head(adj0708) > >> >> MARGIN Poss P235 P247 P703 P218 P430 P489 P83 P307 P337.... > >> >> 1 64.28571 29 0 0 0 0 0 0 0 0 0 > 0 > >> >> 0 0 0 > >> >> 2 -100.00000 6 0 0 0 0 0 0 0 1 0 > 0 > >> >> 0 0 0 > >> >> 3 100.00000 4 0 0 0 0 0 0 0 1 0 > 0 > >> >> 0 0 0 > >> >> 4 -33.33333 7 0 0 0 0 0 0 0 0 0 > 0 > >> >> 0 0 0 > >> >> 5 200.00000 2 0 0 0 0 0 0 0 0 0 > 0 > >> >> -1 0 0 > >> >> 6 -83.33333 12 0 -1 0 0 0 0 0 0 0 > 0 > >> >> 0 0 0 > >> >> > >> >> adj0708 is actually a 35657x341 data set. Each column after > "Poss" > >> is > >> >> an independent variable, the dependent variable is "MARGIN" and > it > >> is > >> >> weighted by "Poss" > >> >> > >> >> > >> >> The regression is below: > >> >> fit.adj0708 <- lm( adj0708$MARGIN~adj0708$P235 + adj0708$P247 + > >> >> adj0708$P703 + adj0708$P430 + adj0708$P489 + adj0708$P218 + > >> >> adj0708$P605 + adj0708$P337 + .... + > >> >> adj0708$P510,weights=adj0708$Poss) > >> >> > >> >> I have two questions: > >> >> > >> >> 1. Is there a way to to condense how I write the independent > >> variables > >> >> in the lm(), instead of having such a long line of code (I have > 339 > >> >> independent variables to be exact)? > >> >> 2. I would like to pair the data to look a regression of the > >> >> interactions between two independent variables. I think it would > >> look > >> >> something like this.... > >> >> fit.adj0708 <- lm( adj0708$MARGIN~adj0708$P235:adj0708$P247 + > >> >> adj0708$P703:adj0708$P430 + adj0708$P489:adj0708$P218 + > >> >> adj0708$P605:adj0708$P337 + ....,weights=adj0708$Poss) > >> >> but there will be 339 Choose 2 combinations, so a lot of > independent > >> >> variables! Is there a more efficient way of writing this code. Is > >> >> there a way I can do this? > >> >> > >> >> Thanks, > >> >> Matt > >> >> > >> >> ______________________________________________ > >> >> R-help@r-project.org mailing list > >> >> https://stat.ethz.ch/mailman/listinfo/r-help > >> >> PLEASE do read the posting guide http://www.R- > project.org/posting- > >> >> guide.html > >> >> and provide commented, minimal, self-contained, reproducible > code. > >> > > > ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
