brianlevey

library(tidyverse)
library(RSQLite)

tmp <- tibble(x = rnorm(100),
              y = 1.8*x + rnorm(100))

mod <- lm(y~x, tmp)
mod$fitted.values <- NULL
mod$residuals <- NULL
mod$effects <- NULL

brianlevey

library(tidyverse)
library(tsibble)
library(feasts)

tmp <- tibble(id = 1:100, x = rnorm(100)) %>%
  mutate(x = cumsum(x),
         y = 1.2 + 2.45*lag(x, 2) + rnorm(100)) %>%
  as_tsibble(index = id)

tmp %>% CCF(x, y) %>%

brianlevey

library(tidyverse)
l = .90 # controls the maximum probability
k = -0.03 # controls the steepnees of the curve
m = 50 # sets the midpoint of the curve

tmp <- tibble(x = 1:200) %>%
    mutate(y = l / (1 + exp(k*(x - m))))

ggplot(tmp) + geom_line(aes(x, y)) +
    geom_vline(xintercept = tmp$x[round(tmp$y, 1)==.5][1], alpha = .5, color = "red") +

brianlevey

cpred <- function(dv, pr, id=NULL, threshold=.5){
  df <- tibble(id, dv, pr) %>%
    mutate(pr = ifelse(pr > threshold, 1, 0)) %>%
    drop_na()
  if(!is.null(id)){
    df <- df %>% group_by(id)
  }
  df <- df %>%
    summarize(P = sum(dv),
              N = sum(dv == 0),

brianlevey

library(tidyverse)

tmp <- tibble(mod = "effect", est = 2.2, se = .75)

ggplot(tmp) + 
    geom_segment(aes(x = est - (2*se), 
                     xend = est + (2*se), 
                     y = mod,
                     yend = mod,
                 color = "gray",

brianlevey

library(tidyverse)
library(broom)

data <- nest(mtcars, -cyl)
data100 <- mutate(mtcars, disp = 100) %>% nest(-cyl, .key = data100)
data200 <- mutate(mtcars, disp = 200) %>% nest(-cyl, .key = data200)

data <- full_join(data, data100, by = "cyl")
data <- full_join(data, data200, by = "cyl")

brianlevey

library(tidyverse)

data <- tibble(id = letters[1:10], x = rnorm(10), y = rnorm(10))

g1 <- ggplot(data) + geom_point(aes(x, y, color = id))


# matrix rotation
theta <- .9
rmat <- matrix(c(cos(theta), -sin(theta),

brianlevey

# sample
n <- 1000
x <- rnorm(n, 25, 1)
y <- rnorm(n, 25.4, 1)


# unpaired
t.test(y, x)

brianlevey

tibble(x = c(0,0,1,0,1,1,1,0,0,1,1,1,1,1,0,0,0)) %>%
  mutate(count = ave(x, cumsum(x==0), FUN=cumsum),
         phase = cumsum(ifelse(count == 1, 1, 0)),
         phase = ifelse(x == 0, 0, phase))

brianlevey

library(tidyverse)

# training data, model has no error
data <- tibble(x0 = 1,
               x1 = rnorm(100),
               y = 0.97 + 1.3*x1)

# ols model
summary(lm(y ~ x1, data))