transition_components.Rmd

---
title: "transition_components"
author: "Anna Quaglieri"
date: "22/11/2018"
output:
  github_document:
    toc: true
---

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

```{r setup1, eval=FALSE}
devtools::install_github("hadley/emo")
devtools::install_github("ropenscilabs/icon")
```


```{r}
library(gganimate)
library(tidyverse)
library(fivethirtyeight)
```


## How does `transition_components()` work?

> ## Transition individual components through their own lifecycle
>
> This transition allows individual visual components to define their own "life-cycle". This means that the final animation will not have any common "state" and "transition" phase as any component can be moving or static at any point in time.

> ## Usage
> transition_components(time, range = NULL, enter_length = NULL,
  exit_length = NULL)

Some key points to keep in mind:

In `transition_components()` you need at least a **time** component and a variable id in your dataset that groups together observations. The `transition_components()` function is useful when you have the same identifier (like a plane, a day, a person, a neighborood etc.) with multiple observations over time.

## A minimal example

The example below is the one provided in the help page `?transition_components` and I added a few simple variants.

* In the standard situation you would use the `group` argument in the `ggplot()` call to define which are the observations with the same IDs.

```{r}
data <- data.frame(
  x = runif(10),
  y = runif(10),
  size = sample(1:3, 10, TRUE),
  time = c(1, 4, 6, 7, 9, 6, 7, 8, 9, 10),
  id = rep(1:2, each = 5)
)

anim <- ggplot(data, aes(x, y, size = size, group = id)) +
  geom_point() +
  transition_components(time)
```


* You can also achieve the same transition but with different colours per ID by using the `colour` argument, which is a more special way of grouping observations.

```{r}
anim <- ggplot(data, aes(x, y, size = size,colour=as.factor(id))) +
  geom_point() +
  transition_components(time)
```


* Notice what happens if you do not set any of the above (`group` or `colour`). You can see that the transition lost the knowledge about what observations belong to the same ID.

```{r}
anim <- ggplot(data, aes(x, y, size = size)) +
  geom_point() +
  transition_components(time)
```

## US births 1994-2003

To show another example, I will use the `US_births_1994_2003` dataset from the `fivethirtyeight` package. The title of the article where this data was used is *Some People Are Too Superstitious To Have A Baby On Friday The 13th*. Is that true?

```{r}
head(US_births_1994_2003)
```

In the example below my **grouping variable** will be Fridays of the month, stored into the variable `date_of_month` after I only filter for Fridays. For example, one id is **1**, which is the 1st Friday of a generic month across years. My plan is to compare the number of babies born on the Fridays 13th of different months (**13**) with babies born on other Fridays! To speed up the process I will compare Fridays 13th with Fridays that occurs on the 1st, 2nd, 3rd, 18th and 28th across months and years.

```{r eval=TRUE}
# Select only Fridays
fridays <- US_births_1994_2003 %>% 
  filter(day_of_week %in% c("Fri") & date_of_month %in% c(1,2,3,13,18,28))

head(fridays)
```

* Grouping using colouring allows me to compare different dates of Fridays

```{r eval=TRUE}
p=ggplot(fridays) + 
  geom_point(aes(x=year,y=births,colour=factor(date_of_month))) +
  theme(legend.position = "bottom") +
  transition_components(time=date)+
  shadow_trail(distance = 0.01, size = 0.3)

animate(p, nframes = 20, 10,duration=10)
```

It looks like between 1994 and 1995 there were consistently less babies born on Fridays 13th, but then this fashion went away. Was that just by chance?!   

## Example with babynames

The `babynames` packages is one of the great packages developed thanks to the effort of another #OzUnconf18 team https://github.com/ropenscilabs/ozbabynames! 

Below I am showing another example that uses `transition_components()` in combination with other fun animations like:

* `shadow_trail()` which allows you to customise the way in which your observations leave a trace of themselves once they move on with their transitions.

* Within `shadow_trail()`, the argument `distance` lets you specify the distance between each trace left. I noticed that it does not work with a very small distance (0.001 wasn't working). It has something to do with the fact that `distance` is used as a denominator at some steps and probably it gets too small.
* The argument `size` works like in the normal `ggplot()` (e.g. size of dots) and it specifies the size of trace left.

```{r}
# install_github("ropenscilabs/ozbabynames")
library(ozbabynames)

p=ggplot(ozbabynames[ozbabynames$name %in% c("Michael","James"),]) + 
  geom_point(aes(x=year,y=count,colour=name)) +
  theme_bw() + 
  transition_components(time=year)+
  shadow_trail(distance = 0.1, size = 2)
p
```

* Let's increase the `distance`

```{r}
p=ggplot(ozbabynames[ozbabynames$name %in% c("Michael","James"),]) + 
  geom_point(aes(x=year,y=count,colour=name)) +
  transition_components(time=year)+
  shadow_trail(distance = 2, size = 2)
p
```

* `distance` too small. The code below will throw the error:

> Error in seq.default(1, params$nframes, by = params$distance) : 
invalid '(to - from)/by'

```{r eval=FALSE}
p=ggplot(ozbabynames[ozbabynames$name %in% c("Michael","James"),]) + 
  geom_point(aes(x=year,y=count,colour=name)) +
  transition_components(time=year)+
  shadow_trail(distance = 0.001, size = 2)

p
```