timcdlucas

#load packages
require(raster)
require(colorplaner)
require(ggplot2)


#here's some dummy data
r1<- raster(ncol=10, nrow=10)
set.seed(0)
values(r1) <- runif(ncell(r1))

timcdlucas

# 1 covariate and intercept
d <- as.matrix(data.frame(intercept = 1, x1 = rnorm(100)))

# Many responses
many_ys <- matrix(d[, 'x1'] * rep(0:10, each = 100) + rnorm(11 * 100, sd = 0.1), ncol = 11)

# Empty list to save our models into.
models <- list()

timcdlucas

library(caret)


m <- train(mpg ~ .,
           data = mtcars,
           method = 'glmnet',
           tuneLength = 10)

plot(m)

timcdlucas

# Assume just the first few parameters vary, but can do the same thing.

ad.surv.intercept.vec <- c(0.0002516054 , 0.0001)
ad.surv.temp.vec <- c(0.1702, 0.2)


tic()
results <- 
  lapply(1:2, function(x) 

timcdlucas

ad.surv.intercept.vec <- c(0.99 * ad.surv.intercept, 1.01 * ad.surv.intercept)
ad.surv.temp.vec <- c(0.99 * ad.surv.temp, 1.01 * ad.surv.temp)

# Going to end up being... 
#   5 runs. 1 with everything at the mean, 2 new ad.surv.intercept and 2 new ad.surv.tmep
N <- 5
df <- data.frame(rep(ad.surv.intercept, N), rep(ad.surv.temp.vec, 5))
df[1:2, 1] <- ad.surv.intercept.vec
df[3:4, 2] <- ad.surv.temp.vec

timcdlucas

# Assume just the first 16 parameters vary, but can do the same thing.
df <-
  data.frame(
    ad.surv.intercept=rep(0.0002516054, 16 * 2),
    ad.surv.temp=0.1702,
    ad.surv.pack=-0.1654,
    disp.intercept=0.0064262,
    disp.pack.size=0.10594,

timcdlucas

# sandbox

library(caret)
library(tictoc)


g1 <- data.frame(mtry = c(2, 5))

tic()

timcdlucas

imp <- list()

for(i in 1:50){
r <- randomForest::randomForest(mpg ~ ., data = mtcars, importance = TRUE)

m <- lm(mpg ~ ., data = mtcars)
m %>% summary

imp[[i]] <- r$importance[,1]

timcdlucas

# Make data with biased train2 set
d <- data.frame(x = runif(300, 0, 5))
d$y <- d$x + rnorm(300)
d$g <- c('train1', 'train2', 'test')
d$y[d$g == 'train2'] <- d$y[d$g == 'train2'] + 5

plot(d$x, d$y, col = factor(d$g))

timcdlucas

library(caret)
library(dplyr)

data(diamonds)


train_i <- sample(c(FALSE, TRUE), nrow(diamonds), replace = TRUE, prob = c(0.2, 0.8))