Broom.Rmd

---
title: "Broom_package"
author: "Rahul Learning"
date: "27 January 2019"
output:
  pdf_document: default
  html_document: default
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

## Broom major functions

1. ###*tidy*: Summarize information about fit components
2. ###*glance*: report goodness of fit measures
3. ###*augment*: add information about observations to a dataset

```{r, message = FALSE}
library(broom)
library(tidyverse)
library(knitr)

```


```{r}
# take a glimpse of the data
glimpse(mtcars)



model1 <- lm(mpg~wt, data = mtcars)

summary(model1)
```

```{r}
# take a glance of the model fitting 

glance(model1)


```

```{r}
# take a glance of the model fitted values and stats

augment(model1) %>% 
  head()


```

```{r, echo = FALSE}
# Print the coefficients
kable2 <- function(data){
  knitr::kable(mutate_if(data, is_numeric, round, 2))
}

tidy(model1) %>% 
  kable2()

```


## Comparing multiple models at once

```{r}

fits <- list(
  fit1 <- lm(hp ~ cyl, data = mtcars),
  fit2 <- lm(hp ~ cyl + mpg, data = mtcars),
  fit3 <- lm(hp ~ ., data = mtcars)
)

gof <- purrr::map_df(fits, glance, .id = "model") %>% 
  arrange(AIC)

gof

```



## Inspecting residuals from multiple linear regression


```{r}

fit <- lm(hp~., data = mtcars)
au <- broom::augment(fit)
au %>% 
  head()


p <- au %>% 
  gather(x, val, -contains(".")) %>% 
  ggplot(aes(x = val, y = .fitted)) +
  geom_point() +
  facet_wrap(~x, scales = "free") +
  labs(x = "Predictpr values", y = "Residuals") +
  theme(axis.text.x = element_blank(),
        axis.ticks.x = element_blank())

p
```


## Bootstrapping examples

```{r}
library(rsample)

boots <- bootstraps(mtcars, times = 100)

boots



fit_nls_on_bootstrap <- function(split){
  nls(               # non linear least square
    mpg ~ k / wt + b,
    analysis(split), # Convert a rsplit object to dataframe
    start = list(k = 1, b = 0)
  )
}

# Fitting model on bootstrap datasets

boot_fits <- boots %>%
  mutate(fit = map(splits, fit_nls_on_bootstrap),
         coef_info = map(fit, tidy))

boot_fits


# Unnesting coefficient info

boot_coefs <- boot_fits %>% 
  unnest(coef_info)


boot_coefs
```


## Plotting uncertainity
```{r}
p <- ggplot(boot_coefs, aes(x = estimate)) +
  geom_histogram(binwidth = 2) +
  facet_wrap(~ term, scales = "free") +
  labs(
    title = "Sampling distribution of k and b",
    y = "count",
    x = "Value"
    )


p



```


## Visualizing uncertainity and predictions


```{r}
boot_aug <- boot_fits %>% 
  mutate(augmented = map(fit, augment)) %>% 
  unnest(augmented)

boot_aug %>% 
  head()

p <- ggplot(boot_aug, aes(x = wt, y = mpg)) +
  geom_point() +
  geom_line(aes(y = .fitted, group = id), alpha = 0.2)

p

```