In amakinney/Stat302Package: What the Package Does (One Line, Title Case)

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(Stat302Package)

This package contains four mathematical functions that calculate t-test values,
  linear regression, and two functions that predict output. To install this
  package from Github, you must first install the package "devtools" in R and
  then in your console, type devtools::install_github("amakinney/Stat302Package")
  then call the package using the library function. 

library(gapminder)
data(gapminder)
my_gapminder <- gapminder
# Tutorial for my_t_test2
my_data <- data.frame(my_gapminder$lifeExp)
x <- as.numeric(my_data[[1]])
alternative <- ("two-sided")
mu <- 60

my_t_test2(x, "two-sided", mu)

my_t_test2(x, "less", mu)

my_t_test2(x, "greater", mu)

The p-value is equal to alpha for the two-sided test, smaller than alpha in the

less than test, and greater than alpha in the greater than test. The only

results that are statistically significant are the second set of tests, where

P is less than alpha.

library(tidyverse)
form <- (lifeExp ~ gdpPercap + continent)
lr <- my_lm(form, my_gapminder)

x <- model.matrix(form, my_gapminder)
beta <- matrix(lr$Estimate)

plott <- x %*% beta


plot(plott)

The gdpPercap coefficient is 4.4527 * 10^-4, so it is almost zero. It shows

that gdpPercap doesn't have much to do with life expectancy.The associated

hypothesis are H0: p = 1/2 and Ha p != 1/2.

The actual versus fitted plot shows that they are actually not too far off the

values, so the model fit would be horizontal through the plot.

library(class)
library(palmerpenguins)
data(penguins)
my_penguins <- penguins

data_clean <- na.omit(my_penguins %>% select(species, body_mass_g, bill_length_mm, 
                                  bill_depth_mm, flipper_length_mm))
data_use <- data_clean %>% select(body_mass_g, bill_length_mm, 
                                  bill_depth_mm, flipper_length_mm)


my_knn_cv(data_use, data_clean$species, k_nn = 1, k_cv = 5)

my_knn_cv(data_use, data_clean$species, k_nn = 5, k_cv = 5)

Based on the CV error, I would be slightly inclined to go with the first model,

with k_nn = 1. The error is only slightly lower, so I don't think I would feel

super strongly either way.

In k-fold cross validation, you are first splitting the data into k parts,

using k-1 folds as the training data to fit the model, using the remaining fold

as test data and make predictions. You then switch which fold is the test data

until every fold has been the test data, then compute the squared error. It is

useful because it ensures each unit of data has been used as test data, so it

it is a very thorough method.

library(randomForest)
library(ggplot2)
set.seed(123)

data <- data.frame(my_penguins)

# Cleans the data
data_cleaned <- my_penguins %>% select(species, body_mass_g, bill_length_mm, 
                                  bill_depth_mm, flipper_length_mm) %>% na.omit()
avg_MSE <- 0

k2 <- my_rf_cv(2)

k5 <- my_rf_cv(5)

k10 <- my_rf_cv(10)

f1 <- ggplot(data = data_cleaned, 
       aes(x = bill_length_mm, y = body_mass_g)) +
  geom_boxplot(fill = "lightgreen") 

f2 <- ggplot(data = data_cleaned, 
       aes(x = bill_depth_mm, y = body_mass_g)) +
  geom_boxplot(fill = "lightgreen")

f3 <- ggplot(data = data_cleaned, 
       aes(x = flipper_length_mm, y = body_mass_g)) +
  geom_boxplot(fill = "lightgreen")

my_table <- matrix(rep(3, times = 6), ncol = 2, byrow = TRUE)
colnames(my_table) <- c("Mean", "SD")
rownames(my_table) <- c("k = 2", "k = 5", "k = 10")
my_table[1,1] = k2
my_table[2,1] = k5
my_table[3,1] = k10
my_table[1,2] = sd(k2)
my_table[2,2] = sd(k5)
my_table[3,2] = sd(k10)
my_table

It appears the smaller k is, the larger the mean is. There is a huge drop

between k = 2 and k = 10. I think this is probably the case because as k

becomes larger, there are more runs through the data so the data becomes

more accurate, and therefore the mean is smaller.

amakinney/Stat302Package documentation built on Dec. 19, 2021, 1:37 a.m.