R: Creating a loop for two models? - r

Question

I currently have following code with two functions that calculate the model fit for two distinct models. The difference is in the lm function, where + log(v2) has been added in model 2.
R code
dat <- data.frame(clicks = runif(30, 1, 100), v1 = runif(30, 1, 100), v2 = runif(30, 1, 100))
p0 <- 1 # number of parameters in lm()
p1 <- 2 # number of parameters in lm()
n <- nrow(dat) - 1
## Model 1 Loop
model1 <- function(x) {
fit <- lm(log(clicks) ~ log(v1), data = dat, subset = 1:x, model = FALSE)
pred <- predict(fit, newdata = dat[x+1, ])
c(summary(fit)$r.squared)
}
## Model 1 Regression
result_m1 <- t(sapply(p0:n, model1))
data.frame(result_m1)
## Model 2 Loop
model2 <- function(x) {
fit <- lm(log(clicks) ~ log(v1) + log(v2), data = dat, subset = 1:x, model = FALSE)
pred <- predict(fit, newdata = dat[x+1, ])
c(summary(fit)$r.squared)
}
## Model 2 Regression
result_m2 <- t(sapply(p1:n, model2))
data.frame(result_m2)
Question: Can I somehow create a function that implements a loop for the different models only, instead of repeating the calculation for every model?
I have something like this in mind but weren't able to implement it .http://www.ats.ucla.edu/stat/r/pages/looping_strings.htm

Answer 1

I don't see a point in recreating a function that can be easily done with model-selection functions in available packages.
library(leaps)
library(dplyr)
b <- regsubsets(clicks ~ ., data=dat, nbest=10, nvmax=2) # carries out exhaustive model selection (10 best models; 2 maximum predictors)
coef(b, 1:3) # returns coefficient for the 3 models in this case
[[1]]
(Intercept) v1
60.8067570 -0.2665699
[[2]]
(Intercept) v2
49.96974177 -0.05227489
[[3]]
(Intercept) v1 v2
62.02323816 -0.26422966 -0.02676747
summary(b)$rsq #provide r.squared value for 3 models
[1] 0.067952759 0.002366681 0.068568059
To run prediction is a tad more complicated.
all.mods <- summary(b)$which[,-1] # gives logic output of predictors combination
all.mods
v1 v2
1 TRUE FALSE
1 FALSE TRUE
2 TRUE TRUE
RHS <- lapply(seq(nrow(all.mods)), function(m) summary(b)$which[m,-1] %>% which %>% names %>% paste(., collapse="+"))
RHS
[[1]]
[1] "v1"
[[2]]
[1] "v2"
[[3]]
[1] "v1+v2"
lm.form <- lapply(RHS, function(m)parse(text=paste("lm(clicks ~", m, ", data=dat)")))
lm.mods <- lapply(lm.form, eval) # return list of all lm.mods generated
The list of lm.mods can subsequently be used for predict with new.data.