r4ds/model-building.Rmd

# Model building

## Introduction

In the previous chapter you learned how linear models worked, and learned some basic tools for understanding what a model is telling you about your data. The previous chapter focussed on simulated datasets to help you learn about how models work. This chapter will focus on real data, showing you how you can progressively build up a model to aid your understanding of the data.

We will take advantage of the fact that you can think about a model partition your data into pattern and residuals. We'll find patterns with visualisation, then make the concrete and precise with a model. We'll them repeat the process, replace the old response variable with the residuals from the model. The goal is to transition from implicit knowledge in the data and your head to explicit knowledge in a quantitative model. This makes it easier to apply to new domains, and easier for others to use. 

For very large and complex datasets this will a lot of work. There are certainly alternative approaches - a more machine learning approach is simply to focus on the predictive ability of the model. These approaches tend to produce black boxes: the model does a really good job at generating predictions, but you don't know why. This is a totally reasonable approach, but it does make it hard to apply your real world knowledge to the model. That, in turn, makes it difficult to assess whether or not the model will continue work in the long-term, as fundamentals change. For most real models, I'd expect you to use some combination of this approach and a more classic automated approach.

It's a challenge to know when to stop. You need to figure out when your model is good enough, and when additional investment is unlikely to pay off. I particularly this quote from reddit user Broseidon241: 

> A long time ago in art class, my teacher told me "An artist needs to know 
> when a piece is done. You can't tweak something into perfection - wrap it up. 
> If you don't like it, do it over again. Otherwise begin something new". Later
> in life, I heard "A poor seamstress makes many mistakes. A good seamstress 
> works hard to correct those mistakes. A great seamstress isn't afraid to 
> throw out the garment and start over."

-- Broseidon241, <https://www.reddit.com/r/datascience/comments/4irajq>

### Prerequisites

We'll start with modelling and EDA tools we needed in the last chapter. Then we'll add in some real datasets: `diamonds` from ggplot2, and `flights` from nycflights13. We'll also need lubridate to extract useful components of datetimes.

```{r setup, message = FALSE}
# Modelling functions
library(modelr)
library(broom)

# EDA tools
library(ggplot2)
library(dplyr)
library(tidyr)

# Data
library(nycflights13)
library(lubridate)
```

## Why are low quality diamonds more expensive?

In previous chapters we've seen a surprising relationship between the quality of diamonds and their price: low quality diamonds (poor cuts, bad colours, and inferior clarity) have higher prices.

```{r dev = "png"}
ggplot(diamonds, aes(cut, price)) + geom_boxplot()
ggplot(diamonds, aes(color, price)) + geom_boxplot()
ggplot(diamonds, aes(clarity, price)) + geom_boxplot()
```

### Price and carat

The basic reason we see this pattern is that the variable that is most predictive of diamond price is its size:

```{r, dev = "png"}
ggplot(diamonds, aes(carat, price)) + 
  geom_hex(bins = 50)
```

To explore this relationship, lets make a couple of tweaks to the diamonds dataset:

1. Remove all diamonds bigger than 2.5 carats
1. Log-transform the carat and price variables

```{r}
diamonds2 <- diamonds %>% 
  filter(carat <= 2.5) %>% 
  mutate(lprice = log2(price), lcarat = log2(carat))

ggplot(diamonds2, aes(lcarat, lprice)) + 
  geom_hex(bins = 50)
```

Log-transforming is very useful here because it make a linear relationship, and linear relationships are generally much easier to work with.  We could go one step further and use a linear model to remove the strong effect of `lcarat` on `lprice`. First, let's see what a linear model tells about the data on the original scale:

```{r}
mod_diamond <- lm(lprice ~ lcarat, data = diamonds2)

grid <- diamonds2 %>% 
  expand(carat = seq_range(carat, 20)) %>% 
  mutate(lcarat = log2(carat)) %>% 
  add_predictions(mod_diamond, "lprice") %>% 
  mutate(price = 2 ^ lprice)

ggplot(diamonds2, aes(carat, price)) + 
  geom_hex(bins = 50) + 
  geom_line(data = grid, colour = "red", size = 1)
```

That's interesting! If we believe our model, then it suggests that the large diamonds we have are much cheaper than expected. This may because no diamond in this dataset costs more than $19,000.

We can also look at the residuals from this model. This verifies that we have successfully removed the strong linear pattern:

```{r}
diamonds2 <- diamonds2 %>% 
  add_residuals(mod_diamond, "lresid")

ggplot(diamonds2, aes(lcarat, lresid)) + 
  geom_hex(bins = 50)
```

Importantly, we can now use those residuals in plots instead of `price`. 

```{r dev = "png"}
ggplot(diamonds2, aes(cut, lresid)) + geom_boxplot()
ggplot(diamonds2, aes(color, lresid)) + geom_boxplot()
ggplot(diamonds2, aes(clarity, lresid)) + geom_boxplot()
```

Here we see the relationship we'd expect. Now that we've removed the effect of size on price,  

To interpret the `y` axis, we need to think about what the residuals are telling us, and what scale they are on. A residual of -1 indicates that `lprice` was 1 unit lower than expected, based on the `carat` alone. $2^{-1}$ is 1/2, so that sugggests diamonds with colour I1 are half the price you'd expect.

### A model complicated model

We could continue this process, making our model complex:

```{r}
mod_diamond2 <- lm(lprice ~ lcarat + color + cut + clarity, data = diamonds2)

add_predictions_trans <- function(df, mod) {
  df %>% 
    add_predictions(mod, "lpred") %>% 
    mutate(pred = 2 ^ lpred)
}

diamonds2 %>% 
  expand_model(mod_diamond2, cut) %>% 
  add_predictions_trans(mod_diamond2) %>% 
  ggplot(aes(cut, pred)) + 
    geom_point()
```


### Exercises

1.  In the plot of `lcarat` vs. `lprice`, there are some bright vertical
    strips. What do they represent?

1.  If `log(price) = a_0 + a_1 * log(carat)`, what does that say about 
    the relationship between `price` and `carat?
    
1.  Extract the diamonds that have very high and very low residuals. 
    Is there any unusual about these diamonds? Are the particularly bad
    or good, or do you think these are pricing errors?

## What affects the number of daily flights?

We're going to start by building a model to help us understand the number of flights per day that leave NYC. We're not going to end up with a fully realised model, but as you'll see, the steps along the way will help us better understand the data.

We'll start by using dplyr to generate the data of interest: 

```{r}
daily <- flights %>% 
  mutate(date = make_datetime(year, month, day)) %>% 
  group_by(date) %>% 
  summarise(n = n())
daily
```

And then we'll plot it with ggplot2: 

```{r}
ggplot(daily, aes(date, n)) + 
  geom_line()
```

### Day of week

This is a small dataset, but there's a lot of pattern to explore. Understanding the long-term trend is challenging because there's a very strong day-of-week effect that dominates the subtler patterns:

```{r}
daily <- daily %>% 
  mutate(wday = wday(date, label = TRUE))
ggplot(daily, aes(wday, n)) + 
  geom_boxplot()
```

There are fewer flights on weekends because a very large proportion of travel is for business. You might sometimes have to less on Sunday for an early flight, but it's very rare that you'd leave on Saturday: you'd much rather be home with your family.

One way to remove this strong pattern is to fit a model that "explains" (i.e. attempts to predict) the day of week effect, and then look at the residuals. Another way of thinking about this is that we're capturing the day-of-week effect, moving it from the data, into a model.

```{r}
mod <- lm(n ~ wday, data = daily)
daily <- daily %>% add_residuals(mod)

daily %>% 
  ggplot(aes(date, resid)) + 
  geom_hline(yintercept = 0, size = 2, colour = "white") + 
  geom_line()

daily %>% 
  expand(wday) %>% 
  add_predictions(mod) %>% 
  ggplot(aes(wday, pred)) + 
    geom_point()
```

Note the change in the y-axis: now we are seeing the deviation from the expected number of flights, given the day of week. This plot is interesting because now that we've removed much of the large day-of-week effect, we can see some of the subtler patterns that remain:

1.  Our day of week adjustment seems to fail starting around June: you can
    still see a strong regular pattern that our model hasn't removed. Drawing
    a plot with one line for each day of the week makes the cause easier
    to see:

    ```{r}
    ggplot(daily, aes(date, resid, colour = wday)) + 
      geom_hline(yintercept = 0, size = 2, colour = "white") + 
      geom_line()
    ```

    The problem appears to be Saturdays: it seems like during summer there are 
    more flights on Saturdays than we expect, and during Fall there are fewer. 
    I suspect this is because of summer holidays: many people go on holiday 
    in the summer, and people don't mind travelling on Saturdays for vacation.
    (This doesn't, however, explain why there are more Satruday flights in 
    spring than fall).

1.  There are some days with far fewer flights than expected:

    ```{r}
    daily %>% filter(resid < -100)
    ```

    If you're familiar with American public holidays, you might spot New Year's 
    day, July 4th, Thanksgiving and Christmas. There are some others that don't 
    seem to correspond immediately to public holidays. You'll work on those 
    in the exercise below.
    
1.  There seems to be some smoother long term trend over the course of a year.
    We can highlight that trend with `geom_smooth()`:

    ```{r}
    daily %>% 
      ggplot(aes(date, resid)) + 
      geom_hline(yintercept = 0, size = 2, colour = "white") + 
      geom_line(colour = "grey50") + 
      geom_smooth(se = FALSE, span = 0.20)
    ```

    There are fewer flights in January (and December), and more in summer 
    (May-Sep). We can't do much more with this trend than brainstorm possible 
    explanations because we only have a single year's worth of data.

### Seasonal Saturday effect

We'll tackle the day of week effect first. Let's zoom in on Saturdays, going back to raw numbers:

```{r}
daily %>% 
  filter(wday == "Sat") %>% 
  ggplot(aes(date, n)) + 
    geom_line() +
    geom_point(alpha = 1/3) + 
    scale_x_datetime(NULL, date_breaks = "1 month", date_labels = "%b")
```

So it looks like summer holidays are from early June to late August. That seems to line up fairly well with the [state's school terms](http://schools.nyc.gov/Calendar/2013-2014+School+Year+Calendars.htm): summer break is Jun 26 - Sep 9. Few families travel in the fall because of the big Thanksgiving and Christmas holidays. So lets add a "term" variable to attemp to control for that. 

```{r}
term <- function(date) {
  cut(date, 
    breaks = as.POSIXct(ymd(20130101, 20130605, 20130825, 20140101)),
    labels = c("spring", "summer", "fall") 
  )
}

daily <- daily %>% mutate(term = term(date)) 

daily %>% 
  filter(wday == "Sat") %>% 
  ggplot(aes(date, n, colour = term)) +
  geom_point(alpha = 1/3) + 
  geom_line() +
  scale_x_datetime(NULL, date_breaks = "1 month", date_labels = "%b")
```

(I manually tweaked the dates to get nice breaks in the plot. Using a visualisation to help you understand what your function is doing is a really powerful and general technique.)

It's useful to see how this new variable affects the other days of the week:

```{r}
daily %>% 
  ggplot(aes(wday, n, colour = term)) +
    geom_boxplot()
```

It looks like there is significant variation across the terms, so fitting a separate day of week effect for each term is reasonable. This improves our model, but not as much as we might hope:

```{r}
mod1 <- lm(n ~ wday, data = daily)
mod2 <- lm(n ~ wday * term, data = daily)

daily %>% 
  gather_residuals(mod1, mod2) %>% 
  ggplot(aes(date, resid, colour = model)) +
    geom_line(alpha = 0.75)
```

That's because this model is basically calculating an average for each combination of wday and school term. (How many observations do we have for each day of week in each term?) We have a lot of big outliers, so they tend to drag the mean far away from the typical value.  We can alleviate this problem by using a model that is robust to the effect of outliers: `rlm`. This greatly reduces the impact of the outliers on our estimates, and gives a result that does a good job of removing the day of week pattern:

```{r, warn = FALSE}
mod3 <- MASS::rlm(n ~ wday * term, data = daily)

daily %>% 
  add_residuals(mod3, "resid") %>% 
  ggplot(aes(date, resid)) + 
  geom_hline(yintercept = 0, size = 2, colour = "white") + 
  geom_line()
```

It's now much easier to see the long-term trend, and the positive and negative outliers.

Very common to use residual plots when figuring out if a model is ok. But it's easy to get the impression that there's just one type of residual plot you should do, when in fact there are infinite.  

### Time of year: an alternative approach

In the previous section we used our knowledge of phenomonen to improve the model. An alternative to using making our knowledge explicit in the model is to give the data more room to speak. We could use a more flexible model and allow that to capture the pattern we're interested in.

When you have a continuous variable in the model, rather than using the unique values that you've seen, it's often more useful to generate an evenly spaced grid. One convenient way to do this is with `modelr::seq_range()` which takes a continuous variable, calculates its range, and then generates an evenly spaced points between the minimum and maximum.

```{r, warn = FALSE}
mod <- MASS::rlm(n ~ wday * yday(date), data = daily)

grid <- daily %>% 
  tidyr::expand(wday, date = seq_range(date, n = 13)) %>% 
  add_predictions(mod)

ggplot(grid, aes(date, pred, colour = wday)) + 
  geom_line() +
  geom_point()
```

(Why use `yday(date)` instead of `date`? That's saying we think that the pattern depends only the day of the year, so we might expect it to be the same in other years. Again, since we only have a single year of data we can't test that hypothesis.)

We know that this pattern doesn't do a good job of capturing the variation in the data. There isn't a simple linear trend across the entire year, so instead we could use a natural spline to allow a smoothly varying trend across the year.

```{r}
library(splines)
mod <- MASS::rlm(n ~ wday * ns(date, 5), data = daily)

daily %>% 
  tidyr::expand(wday, date = seq_range(date, n = 13)) %>% 
  add_predictions(mod) %>% 
  ggplot(aes(date, pred, colour = wday)) + 
    geom_line() +
    geom_point()
```

Particularly, we see the strongly pattern in Saturdays that we identified when coming in the opposite direction. It's always a good sign when you see the same signal from multiple approaches. (But note our previous model was explanatory - this is just predictatory.)

How many degrees of freedom to use? Either pick manually to extract the shape of the data, or you can use one of the model assessment techniques in the following chapter to pick algorithmically. Here we're most interested in explanation, so picking by hand (with a little though and plenty of scepticism) is typically fine.

### Public holidays


### Computed variables

If you're experimenting with many models and many visualisations, it's a good idea to bundle the creation of variables up into a function so there's no chance of accidentally applying a different transformation in different places.

```{r}
compute_vars <- function(data) {
  data %>% mutate(
    term = term(date), 
    wday = wday(date, label = TRUE)
  )
}
```

Another option is to wrap it ito the model formula:

```{r}
wday2 <- function(x) wday(x, label = TRUE)

mod3 <- lm(n ~ wday2(date) * term(date), data = daily)
daily %>% 
  expand(date) %>% 
  add_predictions(mod3)
```

I think this is fine to do provided that you've carefully checked that the functions do what you think they do (i.e. with a visualisation). There are two disadvantages:

1.  You may need to add the variables back in anyway if you want to use
    them in a visualsiation.
    
1.  When looking at the coefficients, their values are longer and harder to 
    read. (But this is a general problem with the way that linear models report
    categorical coefficients in R, not a specific problem with this case.)

### Exercises

1.  Use your google sleuthing skills to brainstorm why there were fewer than
    expected flights on Jan 20, May 26, and Sep 9. (Hint: they all have the
    same explanation.) How would these days generalise to another year?

1.  What do the three days with high positive residuals represent?
    How would these days generalise to another year?

    ```{r}
    daily %>% filter(resid > 80)
    ```

1.  Create a new variable that splits the `wday` variable into terms, but only
    for Saturdays, i.e. it should have `Thurs`, `Fri`, but `Sat-summer`, 
    `Sat-spring`, `Sat-fall`. How does this model compare with the model with 
    every combination of `wday` and `term`?
    
1.  Create a new wday variable that combines the day of week, term 
    (for Saturdays), and public holidays. What do the residuals of 
    that model look like?

1.  What happens if you fit a day of week effect that varies by month?
    Why is this not very helpful? 

1.  Above we made the hypothesis that people leaving on Sundays are more 
    likely to be business travellers who need to be somewhere on Monday.
    Explore that hypothesis by seeing how it breaks down based on distance:
    if it's true, you'd expect to see more Sunday flights to places that
    are far away.

1.  It's a little frustrating that Sunday and Saturday are on separate ends
    of the plot. Write a small function to set the levels of the factor so
    that the week starts on Monday.

1.  Compare the predictions for each `wday` combined with `term` for the 
    `lm` and `rlm`

### Interpolation vs extrapolation

One danger with prediction plots is that it's easy to make predictions that are far away from the original data. This is dangerous because it's quite possible that the model (which is a simplification of reality) will no longer apply far away from observed values. 

As the number of variables in your model grows ...  "the curse of dimensionality": as the number of variables increases the average distance between points increases. That means most of the space is very sparse, and you have to rely on strong assumptions.

To help avoid this problem, it's good practice to include "nearby" observed data points in any prediction plot. These help you see if you're interpolating, making prediction "in between" existing data points, or extrapolating, making predictions about preivously unobserved slices of the data.

One way to do this is to use `condvis::visualweight()`.

<https://cran.rstudio.com/web/packages/condvis/>

## Other types of pattern

Pattern in variance, not just mean.  Standardised residuals.