lm-distcomp-test, R

Language: Layout:

#R version 3.3.2 
# general function for simulating a sample data set given parameters
sim.data <- function(mu, sig, amt, seed, mpar, nl){
  # Simulate data for the practice 
  set.seed(seed)
  x <- rnorm(n=amt, mean=mu, sd=sig)
  
  # create the "true" equation for the regression
  a.true <- mpar[1]
  b.true <- mpar[2]
  y <- x*a.true+b.true
  
  # set the noise level
  noise <- rnorm(n=amt, mean=0, sd=nl)
  d <- cbind(x,y,y + noise)
  colnames(d) <- c("x", "y_true","y")
  return(as.data.frame(d))
}

sim.data1 <-sim.data(10,2,100,2020,c(2,8),1)
sim.data2 <- sim.data(10,2,100,2019,c(2,8),1)
sites <- list(site1 = list(data=sim.data1), site2 = list(data=sim.data2))

head(sim.data1)
head(sim.data2)

# define a residual sum of squares function to handle multiple sites
multi.min.RSS <- function(sites, par){
  rs <- 0
  # calculate the residuals from each data source
  for(site in sites){
    tmps <- par[1] + par[2] * site$data$x - site$data$y
    rs <- rs + sum(tmps^2) #c(rs, tmps)
  }
  # return the square and sum of the residuals
  return(rs)
}

param.fit <- optim(par=c(0,1),
                   fn = multi.min.RSS,
                   hessian = TRUE,
                   sites=sites)
print("Distributed linear model results:")
print(paste("Intercept: ", param.fit$par[1], " Slope: ", param.fit$par[2]))

# stack the data frames vertically for later verification
sim.data3 <- as.data.frame(rbind(sim.data1, sim.data2)) 
print("Base R linear model on the full data set:")
lm(y~x, data=sim.data3)

[ + ] Show input
Absolute running time: 0.24 sec, cpu time: 0.49 sec, memory peak: 32 Mb, absolute service time: 0,25 sec