07-Networks.Rmd

# Networks


**Learning Objectives**

- What is Network data?
- New functions and geoms
- Visualization of nodes and edges as abstract concepts 

## Introduction

This chapter illustrates how to make a Network of data, and how to make practical examples using some of the available packages:

- `{tidygraph}` for Tidy API for Graph Manipulation
- `{ggraph}`    for network visualization
- `{igraph}`    for generating random and regular graphs


## What is network data?

Networks data consists of entities (nodes or vertices) and their relation (edges or links).

Edges can be: directed or undirected


### A tidy network manipulation API

The first package is `tidygraph()` a dplyr API for network data.

New functions:

- `activate()` informs tidygraph on which part of the network you want to work on, either nodes or edges. 
- `.N()` which gives access to the node data of the current graph even when working with the edges
      - `.E()` and `.G()` to access the edges or the whole graph)

```{r 07-01, message=FALSE, warning=FALSE, include=FALSE, paged.print=FALSE}
library(tidyverse)
```


In this **example** we create a graph, assign a random label to the nodes, and sort the edges based on the label of their source node.

The function `play_erdos_renyi()` creates graphs directly through sampling of different attributes.

```{r 07-play_erdos_renyi, message=FALSE, warning=FALSE, paged.print=FALSE}
library(tidygraph)

graph <- tidygraph::play_erdos_renyi(n = 10, p = 0.2) %>% 
  activate(nodes) %>% 
  mutate(class = sample(letters[1:4], n(), replace = TRUE)) %>% 
  activate(edges) %>% 
  arrange(.N()$class[from])

graph
```

### Conversion


Data can be converted with `as_tbl_graph()`, a data structure for tidy graph manipulation. It converts a data frame encoded as an edgelist, as well as converting the result of `hclust()`

```{r 07-highschool}
data(highschool, package = "ggraph")
head(highschool)
```

With `as_tbl_graph()` we obtain:

```{r 07-hs_graph}
hs_graph <- tidygraph::as_tbl_graph(highschool, directed = FALSE)
hs_graph
```


#### hclust() and dist() functions:

In this **example** the `luv_colours()` function allows for all built-in `colors()` translated into **Luv colour space**, a data frame with 657 observations and 4 variables:
[luv_colours](https://github.com/tidyverse/ggplot2/blob/main/data-raw/luv_colours.R)

```{r 07-luv_colours}
luv_colours <- as.data.frame(convertColor(t(col2rgb(colors())),
                                          "sRGB", "Luv"))
luv_colours$col <- colors()
head(luv_colours)
```


This visualization represent the content of the dataset, then we will see how it looks in a grapg representation. 

```{r 07-colors}
ggplot(luv_colours, aes(u, v)) + 
geom_point(aes(colour = col), size = 3) + 
scale_color_identity() + 
coord_equal() +
  theme_void()
```


For example, selecting the first 3 variables and plotting the data with the plot() function we can see that there are some connections within the elements of the dataset, as the colors are connected to each other.
```{r 07-palette}
ggplot2::luv_colours[, 1:3] %>% head
plot(ggplot2::luv_colours[, 1:3])
```


```{r 07-hclust}
luv_clust <- hclust(dist(ggplot2::luv_colours[, 1:3]))
```

```{r 07-class}
class(luv_clust)
```


With the `tidygraph::as_tbl_graph()` function we can transorm the dataset into classes "tbl_graph", "igraph" to make it ready to use for making a visualization of the network data.

```{r 07-luv_graph}
luv_graph <- as_tbl_graph(luv_clust)
luv_graph;class(luv_graph)
```


### Algorithms

The real benefit of networks comes from the different operations that can be performed on them using the underlying structure.

```{r 07-luv_graph2}
luv_graph %>% 
  tidygraph::activate(nodes) %>% 
             mutate(centrality = centrality_pagerank()) %>% 
             arrange(desc(centrality))
```


## Visualizing networks

To visualize the **Network data** we use **{ggraph}**. 

It builds on top of {tidygraph} and {ggplot2} to allow a complete and familiar grammar of graphics for network data. 


### Setting up the visualization

Syntax of **{ggraph}**:

    ggraph() %>%
       ggraph::geom_<functions>

it will choose an appropriate layout based on the type of graph you provide.

[Getting Started guide to layouts](https://ggraph.data-imaginist.com/articles/Layouts.html)

#### Specifying a layout

What is the base requirenment?

The data frame need to be with at least an x and y column and with the same number of rows as there are nodes in the input graph. 

As an **example** we take the `data(highschool, package = "ggraph")` and make a **visualization** of the graph:

      hs_graph <- tidygraph::as_tbl_graph(highschool, 
                                          directed = FALSE)

```{r 07-02, message=FALSE, warning=FALSE, paged.print=FALSE}
library(ggraph)
ggraph(hs_graph) + 
  geom_edge_link() + 
  geom_node_point()
```

A second **example** is with more features:

```{r 07-03}
hs_graph <- hs_graph %>%
  tidygraph::activate(edges) %>% 
             mutate(edge_weights = runif(n()))


ggraph(hs_graph, layout = "stress", weights = edge_weights) + 
  geom_edge_link(aes(alpha = edge_weights)) + 
  geom_node_point() + 
  scale_edge_alpha_identity()
```


In the following **examples** we see different [layouts](https://www.data-imaginist.com/2017/ggraph-introduction-layouts/). 

Information about "drl" type of layout: DRL force-directed graph layout, an be found in the [igraph](https://igraph.org/r/doc/layout_with_drl.html) package.

```{r 07-03b}
layout <- ggraph::create_layout(hs_graph, layout = 'drl')

ggraph(layout) + 
  geom_edge_link() + 
  geom_node_point()
```


Instead of {tidygraph} we use {igraph}, with layout = "kk": layout.kamada.kawai

```{r 07-04, message=FALSE, warning=FALSE, paged.print=FALSE}
require(ggraph)
require(igraph)

hs_graph2 <- igraph::graph_from_data_frame(highschool)

layout <- create_layout(hs_graph2, layout = "kk")

ggraph(layout) + 
    geom_edge_link(aes(colour = factor(year))) + 
    geom_node_point() 
```



A very simple example to understand how to make a graph network is from this tutorial: [Networks in igraph](https://kateto.net/netscix2016.html)

To understand a bit more about the graph structure we can use these functions:

```{r 07-04b}
g1 <- igraph::graph( edges=c(1,2, 2,3, 3, 1), n=3, directed=F ) 


E(g1); # access to the edges
V(g1); # the vertics
g1[] # access to the matrix
```



#### Circularity

Layouts can be **linear** and **circular**.

     coord_polar()   changes the coordinate system and not affect the edges


```{r 07-05}
ggraph(luv_graph, layout = 'dendrogram', circular = TRUE) + 
  geom_edge_link() + 
  coord_fixed()
```


```{r 07-06}
ggraph(luv_graph, layout = 'dendrogram') + 
  geom_edge_link() + 
  coord_polar() + 
  scale_y_reverse()
```

### Drawing nodes


- points
- more specialized geoms: tiles


      geom_node_<functions>
      geom_node_point()
      geom_node_tile() 

[Getting Started guide to nodes](https://ggraph.data-imaginist.com/articles/Nodes.html)



```{r 07-luv_graph_tree}
ggraph(luv_graph, layout = "stress") + 
  geom_edge_link() + 
  geom_node_point(aes(colour =factor(members)),
                  show.legend = F)
```  
    
More features could be added to calculate node and edge centrality, such as:
    
- centrality_power()   
- centrality_degree()  

```{r 07-07}
ggraph(luv_graph, layout = "stress") + 
  geom_edge_link() + 
  geom_node_point(aes(colour =centrality_power()))
```

Or making tiles:

```{r 07-08, message=FALSE, warning=FALSE, paged.print=FALSE}
ggraph(luv_graph, layout = "treemap") + 
  geom_node_tile(aes(fill = depth))
```



### Drawing edges

`geom_edge_link()` draws a straight line between the connected nodes, actually what it does is: it will split up the line in a bunch of small fragments.


- geom_edge_link()
- geom_edge_link2()
- geom_edge_fan()
- geom_edge_parallel()
- geom_edge_elbow()
    - geom_edge_bend() 
    - geom_edge_diagonal()


[Getting Started guide to edges](https://ggraph.data-imaginist.com/articles/Edges.html)

The  `after_stat(index)`:

```{r 07-09}
set.seed(123)
ggraph(hs_graph, layout = "stress") + 
  geom_edge_link(aes(alpha = after_stat(index)))
```


Here is an example about how to use `node.class variable`, the graph is the first that we have seen and it is artificially made with:

        tidygraph::play_erdos_renyi()

```{r 07-10}
graph <- tidygraph::play_erdos_renyi(n = 10, p = 0.2) %>% 
                    activate(nodes) %>% 
                    mutate(class = sample(letters[1:4], 
                                          n(), replace = TRUE)) %>% 
                    activate(edges) %>% 
                    arrange(.N()$class[from])


ggraph(graph, layout = "stress") + 
  geom_edge_link2(
    aes(colour = node.class), 
    width = 3,
    lineend = "round")
```


```{r 07-11}
ggraph(hs_graph, layout = "stress") + 
  geom_edge_parallel()
```

Trees and specifically **dendrograms**:

```{r 07-12}
ggraph(luv_graph, layout = "dendrogram", height = height) + 
  geom_edge_elbow()
```

#### Clipping edges around the nodes

Example: using arrows to show directionality of edges



```{r 07-13}
set.seed(1011)
ggraph(graph, layout = "stress") + 
  geom_edge_link(
    arrow = arrow(), 
    start_cap = circle(5, "mm"),
    end_cap = circle(5, "mm")
  ) + 
  geom_node_point(aes(colour = class), size = 8)
```


#### An edge is not always a line

Nodes and edges are abstract concepts and can be visualized in a multitude of ways.

-  geom_edge_point()


```{r 07-14}
ggraph(hs_graph, layout = "matrix", sort.by = node_rank_traveller()) + 
  geom_edge_point()
```

### Faceting

- facet_nodes() 
- facet_edges()
- facet_graph()


```{r 07-15}
ggraph(hs_graph, layout = "stress") + 
  geom_edge_link() + 
  geom_node_point() + 
  facet_edges(~year)
```


## Conclusions

Making a **{ggraph}** means understanding of the different classes of datasets that can be used inside the function. Also, very important is to have clear in mind the structure of the graph that you would like to acheive for representing your data. 
There are many layouts available, and they differ by the class of provided data.
In addition, to do not forget that you can make a network of data using **{ggplot2}** as well.

### Resources:

- [tidygraph website](https://tidygraph.data-imaginist.com)
- [Data Imaginist](https://ggraph.data-imaginist.com)
- [Imaginist layouts](https://www.data-imaginist.com/2017/ggraph-introduction-layouts/)
- [Network analysis with r](https://www.jessesadler.com/post/network-analysis-with-r/)
- [R and igraph](https://igraph.org/r/)
- [Getting Started guide to layouts](https://ggraph.data-imaginist.com/articles/Layouts.html)
- [Getting Started guide to nodes](https://ggraph.data-imaginist.com/articles/Nodes.html)
- [Getting Started guide to edges](https://ggraph.data-imaginist.com/articles/Edges.html)



## Meeting Videos

### Cohort 1

`r knitr::include_url("https://www.youtube.com/embed/RiACZPhm53Y")`

<details>
  <summary> Meeting chat log </summary>
```
00:09:20	priyanka gagneja:	sorry everyone I just joined
00:09:38	Federica Gazzelloni:	Hello!
00:10:02	priyanka gagneja:	and will probably be a little in and out .. got a not so happy baby today at home
00:18:11	Stan Piotrowski:	I need to take off for a conflict that just came up.  Catch up with you all on slack!
00:20:46	SriRam:	It is the image product ID
00:21:07	SriRam:	All the IDE’s
00:21:50	Kent Johnson:	The IDE codes are defined here: https://ropensci.github.io/bomrang/reference/get_available_imagery.html
00:27:32	SriRam:	The process is called geo-referencing
00:27:58	SriRam:	And image is called a geo-referenced image
00:31:33	SriRam:	Yes, it is a reference system
00:31:38	SriRam:	A coordinate reference
00:33:04	Federica Gazzelloni:	this is the bit that makes the reference: crs = st_crs(sat_vis)
00:49:27	Jiwan Heo:	something just came up, and have to leave. See you all next week!
00:59:58	priyanka gagneja:	I am signing off now , can someone please address and sign off on my behalf. I will send a msg later on slack
```
</details>

`r knitr::include_url("https://www.youtube.com/embed/SV-fN7qYPso")`

<details>
  <summary> Meeting chat log </summary>
```
00:15:36	Ryan S:	https://www.youtube.com/playlist?list=PLkrJrLs7xfbWjD2rp3pIV85lby-tR3Cnu
00:15:48	Lydia Gibson:	Thanks Ryan!
00:16:00	Ryan S:	link to a very good basic tutorial on simple features
00:53:58	Lydia Gibson:	What is GPU?
00:54:23	Ryan S:	GPU is the graphics processing unit (I think)
00:54:31	Lydia Gibson:	Thank you
00:54:32	Ryan S:	it's the part that "draws" on your screen
00:54:43	Lydia Gibson:	Oh okay
00:54:56	Ryan S:	versus the CPU that does calculations
00:55:48	SriRam:	For 2D and non texture plots, I think it is more a RAM issue
00:55:59	Ryan Metcalf:	Oh. I’m so sorry for using Acronyms! Ryan S. is correct. The balance I’m asking Federica is related….”Can I use a slow Laptop or do I have to use a super computer with massive Video card to render these types of graphical objects.
00:56:23	Lydia Gibson:	I always thought CPU was synonymous with computer.
00:57:22	Ryan S:	Ryan, just repurpose the GPUs you currently have that are mining crypto
00:57:38	Ryan Metcalf:	:) Agreed!!!
00:57:57	SriRam:	Lol
00:59:14	SriRam:	If you have a spatial network, do not miss out on “sfnetworks” package
01:00:07	Federica Gazzelloni:	https://kateto.net/netscix2016.html
01:00:14	Federica Gazzelloni:	https://www.data-imaginist.com/2017/ggraph-introduction-layouts/
01:00:24	Federica Gazzelloni:	https://www.hcbravo.org/networks-across-scales/misc/tidygraph.nb.html
01:00:41	Federica Gazzelloni:	https://igraph.org/r/doc/layout_with_drl.html
01:00:48	Federica Gazzelloni:	https://tidygraph.data-imaginist.com/reference/index.html#section-misc
01:00:57	Federica Gazzelloni:	https://ggraph.data-imaginist.com/articles/Layouts.html
01:01:52	Federica Gazzelloni:	https://web.stanford.edu/class/bios221/book/Chap-Graphs.html
https://github.com/jtichon/ModernStatsModernBioJGT/tree/master/data
https://simplemaps.com/data/world-cities
01:07:00	SriRam:	Tidy is I think , Hadley definition, variable is a column, sample point is a row
01:07:36	SriRam:	Sorry my microphone does not work since a few sessions now 🙁
01:07:50	Ryan S:	borders on "marketing" to some degree.  :)
01:08:08	SriRam:	Sfnetwork is not for graphs, it is more for spatial operations
```
</details>