hi all - quick question:
i have a matrix 'y' of response values, with two explanatory variables
'x1' and 'x2'.
tested values of 'x1' and 'x2' are sitting in two vectors 'x1' and
'x2'.
i want to learn model parameters without "unrolling" the matrix of
response values.
example below:
# some fake data for the example
x1 <- 1:5
x2 <- 1:10
y <- matrix(runif(50), nrow = 5)
# current method:
z <- vector()
for(i in x1) for(j in x2) z <- c(z, i, j, y[i, j])
z <- data.frame(matrix(z, ncol = 3, byrow = TRUE))
colnames(z) <- c("x1", "x2", "y")
m <- glm(y ~ x1 + x2 + x1:x2, family = binomial, data = z)
# what i'd like to do, kind of:
m <- glm(y ~ x1 + x2 + x1:x2)
basically, i have to "unfold" the matrix 'y' to a data frame 'z' then
solve.
this is somewhat tedious.
anyone know of a way i can do this more generally, especially if
working in even higher dimensions than 2 (i.e. with an arbitrary-
dimension array of response values)?
fitting a glm with matrix of responses
2 messages · Murat Tasan, David Winsemius
On Nov 11, 2009, at 2:24 PM, Murat Tasan wrote:
hi all - quick question:
i have a matrix 'y' of response values, with two explanatory variables
'x1' and 'x2'.
tested values of 'x1' and 'x2' are sitting in two vectors 'x1' and
'x2'.
i want to learn model parameters without "unrolling" the matrix of
response values.
example below:
# some fake data for the example
x1 <- 1:5
x2 <- 1:10
y <- matrix(runif(50), nrow = 5)
# current method:
z <- vector()
for(i in x1) for(j in x2) z <- c(z, i, j, y[i, j])
z <- data.frame(matrix(z, ncol = 3, byrow = TRUE))
colnames(z) <- c("x1", "x2", "y")
m <- glm(y ~ x1 + x2 + x1:x2, family = binomial, data = z)
# what i'd like to do, kind of:
m <- glm(as.vector(y) ~ expand.grid(x1 , x2)^2)
Perhaps:
> zdf <- expand.grid(x1,x2)
> zdf$y <- as.vector(y)
> m <- glm(as.vector(y) ~ (Var1 + Var2)^2, data=zdf)
> m
Call: glm(formula = as.vector(y) ~ (Var1 + Var2)^2, data = zdf)
Coefficients:
(Intercept) Var1 Var2 Var1:Var2
0.425943 0.066960 -0.001198 -0.006480
Degrees of Freedom: 49 Total (i.e. Null); 46 Residual
Null Deviance: 4.067
Residual Deviance: 3.759 AIC: 22.5
basically, i have to "unfold" the matrix 'y' to a data frame 'z' then solve.
as.vector would do that, but I don't know how well it works within a formula.
this is somewhat tedious. anyone know of a way i can do this more generally, especially if working in even higher dimensions than 2 (i.e. with an arbitrary- dimension array of response values)?
?formula You can get all two way interactions with ( )^2 David Winsemius, MD Heritage Laboratories West Hartford, CT