03_preprocessing.Rmd

# Data Preprocessing

While exploring data, we need to preprocess it properly for our purpose.
And these work can be repeat until we construct a dataset that can make our probabilistic model well.

## ifelse

`ifelse` is `Conditional Statments function` that is very simple and useful in R. We can use this with just 3 parameters - condition, return value when it's true, return value when it's false.

let's see example code.

first, print the head of dataset
```{r}
head(mpg$hwy)
```

And let's apply ifelse to this head.
if the hwy(highway fuel efficiency) is higher than 30, returns 'Good'.
```{r}
ifelse(head(mpg$hwy) > 30, 'Good', 'Bad')
```

We can use this data by assign method(`<-`). Let's Control Entire variable.

```{r}
evaluate_hwy <- ifelse(mpg$hwy > 30, 'Good', 'Bad')
print(evaluate_hwy)
```

Then, Why is this necessary?

Sometimes we need categorical data(nominal data or ordinal data) for not only visualization but also data handling. Especially, We do not always get gaussian distribution dataset, so since the data is not uniform, it is necessary to divide it into appropriate intervals.

We can use this function to not only create new variable but also add another column.
```{r}
df_mpg <- mpg
df_mpg['eval_hwy'] <- ifelse(df_mpg$hwy>30, 'High', 'Low')
df_mpg$eval_hwy
```

## dplyr :: mutate

In R, `dplyr` package is the most popular library for data preprocessing. Especially it provides pipe('%>%'), and it's very helpful for intuitive coding. And `mutate` in `dplyr` package do exactly same work as the one above, However it is simpler and easier to use. Let's see this.

(before load, you have to install "dplyr" package)
```{r}
# install.packages("dplyr") 
library(dplyr)
```

```{r}
head(df_mpg)
```
We'll use `mutate` function and `pipe` syntex to handle `cty`(city fuel economy) column.

```{r}
df_mpg %>%
  mutate(mean_fuel = (cty + hwy)/2) %>%
  head
```

We can see the last column is added and named with `mean_fuel`. We didn't have to use the name of dataframa(df_mpg) repeatly to use columns. It's very nice.

And of course we can use `ifelse` function with the `pipe` and `mutate`.

```{r}
df_mpg %>%
  mutate(mean_fuel = (cty + hwy)/2, 
         eval_fuel = ifelse(mean_fuel > 20, 'high', 'low')) %>%
  head
```

If you want to use this data table continuously, you have to assign this table to the new variable like this.

```{r}
new_mpg <- df_mpg %>%
            mutate(mean_fuel = (cty + hwy)/2, 
            eval_fuel = ifelse(mean_fuel > 20, 'high', 'low'))
```

```{r}
str(new_mpg)
```

```{r}
qplot(new_mpg$mean_fuel)
```

```{r}
qplot(new_mpg$eval_fuel)
```

We can use more functions from `dplyr` package. The commonly used functions that except mutate we have seen already are `filter()`, `select()`, `arrange()`, `summarise()`. Let's look at these one by one.

## dplyr :: filter

First, `filter()` function. `filter` extracts rows by some conditions. 
```{r}
head(new_mpg)
```
If you want to extract only `audi`, you can use `filter`.

```{r}
new_mpg %>%
  filter(manufacturer=='audi')
```

If you want to extract `audi a4` model, 
```{r}
new_mpg %>%
  filter(manufacturer=='audi' & model=='a4')
```

If you want to extract `audi a4` that have `Good for highway fuel economy`,
```{r}
new_mpg %>%
  filter(manufacturer=='audi' & model=='a4' & eval_hwy == 'High')
```
Yes, now you can choose the car that you want.

## dplyr :: select

Next, `select()` function. 
We have used `filter()` to extract rows, now we'll use `select()` to extract columns.

```{r}
new_mpg %>%
  select(manufacturer) %>%
  head
```

Two or more columns can be extracted at the same time.

```{r}
new_mpg %>%
  select(manufacturer, class, hwy) %>%
  head
```


And with pipe, we can extract some rows and some columns at the same time too.

```{r}
new_mpg %>%
  select(manufacturer, model, class, hwy) %>%
  filter(hwy > 23) %>%
  head(10)
```

## dplyr :: arrange

In case of numerical data(variable), we can order it. If data is categorical variable, It'll return error.

```{r}
new_mpg %>%
  select(manufacturer, model, class, hwy) %>%
  arrange(hwy) %>%
  head(10)
```

If you want to sort dataset from highest to lowest highway fuel economy, wrapping arrange variable by `desc()`

```{r}
new_mpg %>%
  select(manufacturer, model, class, hwy) %>%
  arrange(desc(hwy)) %>%
  head(10)
```

## dplyr :: summarise

The function `summarise()` can be used to get basic statistics by specifying functions such as mean(), sd(), median(), etc.

```{r}
new_mpg %>%
  filter(manufacturer=='audi') %>%
  summarise(mean_hwy = mean(hwy))
  head(10)
```

```{r}
new_mpg %>%
  filter(manufacturer=='audi') %>%
  summarise(mean_hwy = mean(hwy),
            max_hwy = max(hwy),
            min_hwy = min(hwy))
  head(10)
```
But, this function can be more useful if used together with groupby.

```{r}
new_mpg %>%
  group_by(manufacturer) %>%
  summarise(mean_hwy = mean(hwy),
            max_hwy = max(hwy),
            min_hwy = min(hwy))
```

we can apply `arrange()` too.
```{r}
new_mpg %>%
  group_by(manufacturer) %>%
  summarise(mean_hwy = mean(hwy),
            max_hwy = max(hwy),
            min_hwy = min(hwy)) %>%
  arrange(desc(mean_hwy))
```

We select 2 or more columns to `group_by` function like this.

```{r}
new_mpg %>%
  group_by(manufacturer, model) %>%
  summarise(mean_hwy = mean(hwy),
            max_hwy = max(hwy),
            min_hwy = min(hwy))
```

and apply `filter()` too.

```{r}
new_mpg %>%
  group_by(manufacturer, model) %>%
  summarise(mean_hwy = mean(hwy),
            max_hwy = max(hwy),
            min_hwy = min(hwy)) %>%
  filter(manufacturer=='audi' | manufacturer=='hyundai')
```


## dplyr :: left_join

`dplyr` package contain `join` functions. First, we try `left_join()` for our custom data.


```{r}
students <- c('Jennie', 'Tom', 'Minsu', 'Jay', 'Bob')
classes <- c('A','B','C','D','E')
keys <- c('JN', 'T', 'M', 'J', 'B')
heights <- c(155, 190, 165, 177, 180)
weights <- c(57, 101, 64, 80, 88)

class_info <- data.frame(students, classes)
students_info <- data.frame(students, keys, heights, weights)

head(class_info)
head(students_info)
```

If we have to combine this two table, We need some "point"s. This means some "column"s that two table have in common. In this example, `students` column is that.

Let's combine datasets by Setting `students` column as a point.
```{r}
all_info <- left_join(class_info, students_info, by='students')
all_info
```


## dplyr :: right_join

What if we use `right_join()`?
In this table, it's same.

```{r}
all_info2 <- right_join(class_info, students_info, by='students')
all_info2
```

Because left table(class_info) and right rable(students_info) have same variable(students).

If one side has Na in `students` column, the result will be different.

```{r}
students_info2 <- students_info[-1,]
students_info2
```

```{r}
all_info3 <- left_join(class_info, students_info2, by='students')
all_info3
```

let's see this. We have founded 3 NA values. It's bacause students_info2 didn't have the Jennie's informatino(row)
But,
```{r}
all_info4 <- right_join(class_info, students_info2, by='students')
all_info4
```

If we have chosen `right_join()` at this moment, we can get different result like this.
It's because students_info2 didn't have the Jennie's informatino(row). Same.

`right_join()` receives two inputs like `left_join()`, and joins tables based on the right (second) input.


## dplyr :: inner_join

How can we join tables if we met all of tables are have `NA` at base column of reference?
Like this case,
```{r}
class_info2 <- class_info[-2,]
class_info2
```

```{r}
all_info4 <- left_join(class_info2, students_info2, by='students')
all_info4
```

```{r}
all_info5 <- right_join(class_info2, students_info2, by='students')
all_info5
```

Yes, returns have `NA` all as expected.
But we want to get result that without missing value. In this case, we can use `inner_join()`
```{r}
all_info6 <- inner_join(class_info2, students_info2, by='students')
all_info6
```


## dplyr :: full_join

Conversely, if we want to display all values together even if there are missing values in left or right, we can use `full_join()`

```{r}
all_info7 <- full_join(class_info2, students_info2, by='students')
all_info7
```


## dplyr :: bind_cols
If two data tables have the same number of rows, you can use `bind_cols()`. It is simple and return different result from `join`.
```{r}
bind_cols(class_info2, students_info2)
```

Yes, it just concatenate two tables. There is no reference column.
So, same number of rows should be noted. If each number of rows are different, It returns error.

```{r}
class_info
```

```{r}
students_info2
```

```
bind_cols(class_info, students_info2)
```

```
## Error: Can't recycle `..1` (size 5) to match `..2` (size 4). Run `rlang::last_error()` to see where the error occurred.
```


## dplyr :: bind_rows

Unlike `bind_cols()`, it can join tables with different columns.


```{r}
fruit <- c('apple', 'banana', 'watermelon', 'mango')
berry <- c('strawberry', 'raspberry', 'mulberry', 'wildberry')

print(fruit)
print(berry)
```



```{r}
fruit_df <- data.frame(fruit)
fruit_df
```


```{r}
berry_df <- data.frame(berry)
berry_df
```

```{r}
fruit_df$sugar <- c(3,8,2,9)
berry_df$sugar <- c(2,1,1,1)
berry_df$color <- c('red', 'red', 'black', 'red')

print(fruit_df)
print(berry_df)
```


```{r}
berry_df <- rename(berry_df, fruit = berry)
print(berry_df)
```


```{r}
bind_rows(fruit_df, berry_df)
```

Yes, It returns data table without error even if the structure of column is different.
And of course, the number of rows doesn't matter, since we're joining rows. But let's check them too.

```{r}
bind_rows(fruit_df, berry_df[-1,])
```

Okay, Just one row decreased.

## reshape :: melt

`reshape` is the most representative r package for data restructuring.

We'll cover two functions that make the reshape package useful. One `melt()` that melts (deconstructs) the data, the other `cast()` that casts (reconstructs) the data. As you can see from the word, the name is compared to the casting process of iron.

```{r}
head(airquality)
```

```{r}
library(reshape)
sample_decon_df <- melt(airquality, id=c('Month','Day'))
head(sample_decon_df)
```

```{r}
summary(sample_decon_df$variable)
```
We can see that all columns except Month and Day, specified as id values in the melt function, are expressed as `variable`.


## reshape :: cast

The `cast()` function can cast the data melted above. Here, R's `fomula` should be handled flexibly, and you can learn it by implementing it in code.

> 
Fomula means an expression that expresses a function like a regular expression in text terms, and it is good to learn '~' and '+' at the most basic level.

```{r}
head(cast(sample_decon_df, Month+Day~variable))
```

You can see that the melted data is cast. There is a difference from the original, in that an index is added and the position of the column is changed. It's not that important, so let's move on.

In formula, we defined to use both Month and Day as dependent variables through the '+' sign, and defined variable as the independent variable through the '~' sign.

More precisely, what comes before the '~' sign is the dependent variable, and what comes after the sign is the independent variable. This can be thought of as f(x)~x.

Let's try more.

```{r}
cast(sample_decon_df, Month~variable)
```

Only month was used as the dependent variable. The explanation of the `fomula` above will make more sense.

```{r}
cast(sample_decon_df, Month~Day+variable)[1:4,1:5]
```

This time, Day was added to the independent variable, not the dependent variable. It's a little complicated, but if you understand it, you can see that it is possible to further subdivide each independent variable into days, such as ozone on day 1 and solar on day 1.

## reshape :: cast with aggregation

Cast also has data aggregation capabilities. It is very simple and powerful. let's try.

```{r}
head(cast(sample_decon_df, Month+Day~variable, mean))
```

It doesn't look any different from the original. This is because there is only one data value for each date, so it cannot be aggregated. To be precise, the aggregation for a day is the same as the data for that day.

Except for Day, you can look at monthly data like this.

```{r}
cast(sample_decon_df, Month~variable, mean)
```

Wait, if there is NA in some data like this, it is displayed as NA because aggregation is impossible. To prevent this we can set `na.rm` to True.
  
```{r}
cast(sample_decon_df, Month~variable, mean, na.rm=T)
```

Of course, it is possible to count other than average.

```{r}
cast(sample_decon_df, Month~variable, max, na.rm=T)
```
```{r}
cast(sample_decon_df, Month~variable, min, na.rm=T)
```