Thanks for getting back to me so quickly greg. Im not quite sure how to do what you just said, is there an example that you can show?
I understand how to create the string with a formula in it but im not sure how to loop through the pairs of variables? How do I first get these 2way interaction variables, I can no longer use the "^" right? Sorry for so many questions, Matt On Thu, Mar 3, 2011 at 4:16 PM, Greg Snow <greg.s...@imail.org> wrote: > 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.