---
title: "Supervised learning exercises"
author: "Nicholas Horton (nhorton@amherst.edu)"
date: "July 25, 2017"
output:
  html_document:
    fig_height: 5
    fig_width: 7
  pdf_document:
    fig_height: 5
    fig_width: 7
  word_document:
    fig_height: 3
    fig_width: 5
---

```{r, setup, include=FALSE}
library(mdsr)   # Load additional packages here 


# Some customization.  You can alter or delete as desired (if you know what you are doing).
trellis.par.set(theme=theme.mosaic()) # change default color scheme for lattice
knitr::opts_chunk$set(
  tidy=FALSE,     # display code as typed
  size="small")   # slightly smaller font for code
```

## Introduction
These exercises are taken from the supervised learning chapter from **Modern Data Science with R**: http://mdsr-book.github.io.  Other materials relevant for instructors (sample activities, overview video) for this chapter can be found there.


## Sleep
The ability to get a good night's sleep is correlated with many positive health outcomes. The `NHANES` data set in the `NHANES` package contains a binary variable `SleepTrouble` that indicates whether each person has trouble sleeping. For each of the following models:

1. Build a classifier for `SleepTrouble`
2. Report its effectiveness on the `NHANES` training data
3. Make an appropriate visualization of the model
4. Interpret the results. What have you learned about people's sleeping habits?
  
You may use whatever variable you like, except for `SleepHrsNight`.
Models:

- Null model
- Logistic regression
- Decision tree
- Random forest
- Neural network
- Naive Bayes
- K nearest neighbors

SOLUTION:

```{r}
library(mdsr)   
library(NHANES)
# solution goes here
```

## Quantitative sleep
Repeat the previous exercise, but now use the quantitative response variable `SleepHrsNight`. Build and interpret the following models:

- Null model
- Multiple regression
- Regression tree
- Random forest
- Ridge regression
- LASSO

SOLUTION:

```{r}
library(mdsr)   
library(NHANES)
# solution goes here
```

## Even more sleep
Repeat either of the previous exercises, but this time first separate the 
`NHANES` data set uniformly at random into 75% training and 25% testing sets. Compare the effectiveness of each model on training vs. testing data. 

SOLUTION:

```{r}
library(mdsr)   
library(NHANES)
# solution goes here
```

## Pregnant?
Repeat the first exercise, but for the variable `PregnantNow`. What did you learn about who is pregnant? 

SOLUTION:

```{r}
library(mdsr)   
library(NHANES)
# solution goes here
```

## NASA weather
The `nasaweather` package contains data about tropical `storms` from 1995-2005. Consider the scatterplot between the `wind` speed and `pressure` of these `storms` shown below.

<<storms, message=FALSE>>=
library(mdsr)
library(nasaweather)
ggplot(data = storms, aes(x = pressure, y = wind, color = type)) + 
  geom_point(alpha = 0.5)
@

The `type` of storm is present in the data, and four types are given: extratropical, hurricane, tropical depression, and tropical storm. There are [complicated and not terribly precise](https://en.wikipedia.org/wiki/Tropical_cyclone#Classifications.2C_terminology.2C_and_naming) definitions for storm type. Build a classifier for the `type` of each storm as a function of its `wind` speed and `pressure`.  

Why would a decision tree make a particularly good classifier for these data? Visualize your classifier in the data space in a manner 
similar to Figure 8.10 or 8.11.  

SOLUTION:

```{r message=FALSE}
library(mdsr)   
library(nasaweather)
# solution goes here
```


## Arrival delays
Fit a series of supervised learning models to predict arrival delays for flights from 
New York to `SFO` using the `nycflights13` package.  How do the conclusions change from
the multiple regression model presented in the Statistical Foundations Chapter?

SOLUTION:

```{r}
library(mdsr)   
library(nasaweather)
# solution goes here
```

Use the College Scorecard Data (https://collegescorecard.ed.gov/data) to model student debt as a function of institutional characteristics using the techniques described in this chapter.  Compare and contrast results from at least three methods.  (Note that a considerable amount of data wrangling will be needed.)

SOLUTION:

```{r}
library(mdsr)   
# solution goes here
```