knitr::opts_chunk$set( collapse = TRUE, comment = "#>" )  library(DiagrammeR)  Occasionally, you'll want to operate on a select group of nodes or edges. Some functions affect a single node or edge while others (or, sometimes, the same functions) operate on all nodes or edges in a graph. Selections allow you to target specified nodes or edges and then apply specialized functions to operate on just those selected entities. Most of the selection functions support rudimentary set operations across several calls of the selection functions (i.e., for the union, intersection, or difference between selection sets of nodes or edges). # Creating a Node Selection Selecting nodes in a graph is accomplished by targeting a specific node attribute (e.g., type, label, styling attributes such as width, if available, or arbitrary data values available for nodes). As with all of the select_*() functions, the graph argument is the first in the function signature. This is beneficial for forward-piping the graph in a pipeline and propagating a series of tranformations on the graph. There may be situations where several types of selections be applied to the graphs nodes or edges (say, selecting all nodes of a graph in the first step and then subtracting nodes of a specific type from that group) and this is where the pipeline paradigm becomes incredibly useful, and, as an added bonus, easy to read and to reason about. The main handle on selectivity for the select_nodes() function is through the node attributes of the graph's internal NDF, which contains columns for all possible node attributes in the graph. By providing an expression in conditions, nodes will be placed in the active selection where they are TRUE. Here are two common types of expressions that work well for the conditions argument: 1. a logical expression with a comparison operator (>, <, ==, or !=) 2. a regular expression for filtering via string matching Let's create a simple graph with four nodes that have numeric data values. Then, we'll inspect the graph's internal NDF to see where we are starting from. # Create a node data frame ndf <- create_node_df( n = 4, data = c( 9.7, 8.5, 2.2, 6.0) ) # Create a new graph based on the ndf graph <- create_graph(nodes_df = ndf) # Inspect the graph's NDF graph %>% get_node_df()  In the first example, nodes will be selected based on a logical expression operating on a collection of numeric values. To examine the result of the selection, the get_node_df_ws() will be used. This function will display a table for the active selection of nodes in the graph (like get_node_df(), but with a subset of nodes). # Select nodes where the data attribute # has a value greater than 7.0 (it's the # first 2 nodes) graph <- graph %>% select_nodes( conditions = data > 7.0 ) # Get the graph's current selection # of nodes as a table graph %>% get_node_df_ws()  A selection of nodes can be obtained through a match using a regular expression operating on a collection of character-based values (the node attribute that is named fruits). This uses the grepl() function in the expression supplied to conditions. The regular expression to use is ^ap, where the ^ denotes the beginning of the text to the parsed. # Create a node data frame ndf <- create_node_df( n = 4, fruits = c( "apples", "apricots", "bananas", "plums") ) # Create a new graph based on the ndf graph <- create_graph(nodes_df = ndf) # Select nodes where the fruits # attribute has a match on the first # letters being ap (the first 2 nodes) graph <- graph %>% select_nodes( conditions = grepl("^ap", fruits) ) # Get the graph's current selection # of nodes as a table graph %>% get_node_df_ws()  The situation may arise when a more specialized match needs to be made (i.e., matching this but not that, or, matching two different types of things). This is where the set_op argument can become useful. When a selection of nodes is obtained using select_nodes() (or any of the other select_*() functions that operate on nodes), the selection is stored in the graph object. This is seen in the above examples where get_selection() was used to verify which nodes were in the selection. Because the selection is retained (at least until clear_selection() is called, or, a selection of edges is made), multiple uses select_nodes() can modify the set of selected nodes depending on the option provided in the set_op argument. These set operations are: • union: creates a union of selected nodes in consecutive operations that create a selection of nodes (this is the default option) • intersect: modifies the list of selected nodes such that only those nodes common to both consecutive node selection operations will retained • difference: modifies the list of selected nodes such that the only nodes retained are those that are different in the second node selection operation compared to the first These set operations behave exactly as the base R functions: union(), intersect()/intersection(), and setdiff() (which are actually used internally). Furthermore, most of the select_*() functions contain the set_op argument, so, they behave the same way with regard to modifying the node or edge selection in a pipeline of selection operations. As examples are important in fully understanding how these can work for more complex selections, quite a few will be provided here. The example graph will now be a bit more complex. It will contain 9 nodes, of three different types (fruit, veg, and nut). The label node attribute has the name of the food, and the count attribute contains arbitrary numeric values. # Create a node data frame ndf <- create_node_df( n = 9, type = c( "fruit", "fruit", "fruit", "veg", "veg", "veg", "nut", "nut", "nut"), label = c( "pineapple", "apple", "apricot", "cucumber", "celery", "endive", "hazelnut", "almond", "chestnut"), count = c( 6, 3, 8, 7, 2, 6, 9, 9, 7) ) # Create a new graph based on the ndf graph <- create_graph(nodes_df = ndf) # Inspect the graph's NDF graph %>% get_node_df()  Let's successively use two select_nodes() calls to select all those foods that either begin with c or end with e. Because this is an OR set, we will use the union set operation in the second call of select_nodes(). # Select all foods that either begin # with c or ending with e graph_1 <- graph %>% select_nodes( conditions = grepl("^c", label) ) %>% select_nodes( conditions = grepl("e$", label),
set_op = "union"
)

# Get the graph's current selection
# of nodes as a table
graph_1 %>% get_node_df_ws()


The conditions don't need to be related. In the first below, it's a matching expression using grepl(). The second is filtering to those nodes with count < 5.

# Select any food beginning with a and
# having a count less than 5
graph_2 <-
graph %>%
select_nodes(
conditions = grepl("^a", label)
) %>%
select_nodes(
conditions = count < 5,
set_op = "intersect"
)

# Get the graph's current selection
# of nodes as a table
graph_2 %>% get_node_df_ws()


The following example contains a pair of select_nodes() statements, where the first filters nodes to a fruit group (type == "fruit") and then excludes a subset of these nodes (any fruit containing "apple" in the name) using the set_op = "difference".

# Select any fruit not containing
# apple in its name
graph_3 <-
graph %>%
select_nodes(
conditions = type == "fruit"
) %>%
select_nodes(
conditions = grepl("apple", label),
set_op = "difference"
)

# Get the graph's current selection
# of nodes as a table
graph_3 %>% get_node_df_ws()


There is an additional filtering option available as the nodes argument. Here, a vector of node ID values can be supplied and this will indicate to the function that only that subset of nodes will be considered for select_nodes(). Note that, if nothing is provided in conditions and nodes is given a vector of node ID values, it will be those very nodes that will make up the selection in this function call. While this is convenient and often a good method for selecting nodes (so long as one knows which node IDs need to be selected), the function select_nodes_by_id() handles this use case more directly (as it only filters based on its nodes argument).

# Create a node data frame
ndf <-
create_node_df(
n = 10,
data = seq(0.5, 5, 0.5)
)

# Create a new graph based on the ndf
graph <- create_graph(nodes_df = ndf)

# Inspect the graph's NDF
graph %>% get_node_df()


Let's now perform a call to select_nodes() that contains both an expression for the conditions argument, and, a range of node ID values in the nodes argument.

# Select from a subset of nodes
# (given as nodes = 1:6) where
# the data value is greater than 1.5
graph <-
graph %>%
select_nodes(
conditions = data > 1.5,
nodes = 1:6
)

# Get the graph's current selection
# of nodes as a table
graph %>% get_node_df_ws()


# Creating an Edge Selection

Selecting edges in a graph is done in a manner quite similar to selecting nodes. The primary means for targeting a specific edges is through any available edge attributes (e.g., rel, styling attributes such as color, or arbitrary data values available for edges).

We will start by creating a graph with four nodes and four edges. The edges will have an edge attribute called data that contains numeric values.

# Create a node data frame
ndf <- create_node_df(n = 4)

# Create an edge data frame
edf <-
create_edge_df(
from = c(1, 2, 3, 4),
to = c(2, 3, 4, 1),
data = c(
8.6, 2.8, 6.3, 4.5)
)

# Create a new graph from
# the NDF and EDF
graph <-
create_graph(
nodes_df = ndf,
edges_df = edf
)

# Inspect the graph's EDF
graph %>% get_edge_df()


The following example shows how to create selections of edges based on a logical expression operating on a collection of numeric values in the data attribute.

# Select edges where the data
# attribute has a value
# greater than 5.0
graph <-
graph %>%
select_edges(
conditions = data > 5.0
)

# Get the graph's current selection
# of edges as a table
graph %>% get_edge_df_ws()


# Selecting the Last Node or Edge in an NDF or EDF

You can select the last node or edge from the graph's internal node data frame (NDF) or internal edge data frame (EDF), respectively. Usually, this will be the last node or edge created since new nodes or edges are added to the bottom of the data frame and there is no shuffling of these positions. Immediately after creating a single node or edge, calling either the select_last_node() or the select_last_edge() functions will result in a selection of the last node or edge created.

For both functions, graph is the only argument. If we provide a graph object to select_last_nodes_created() we will get an active selection of nodes (the last nodes that were added to the graph object). Likewise, the select_last_edges_created() will create an active selection of edges, whichever edges were last defined in a function call.

To begin, we'll use the same graph as before but we'll clear any active selection using the clear_selection() function. To ensure that there is no active selection of nodes or edges, we can use the get_selection() function. A value of NA means that there is neither a selection of nodes nor edges.

# Clear the graph's selection
graph <-
graph %>%
clear_selection()

# Check whether there is still
# a selection present
graph %>% get_selection()


If we use select_last_nodes_created on this graph, we find that the active selection of nodes contains all the nodes in the graph, as they were all produced in the create_graph() function (and no nodes were added in subsequent operations).

# Select the last node in the graph's NDF and confirm
# that the selection was made
graph %>%
select_last_nodes_created() %>%
get_node_df_ws()


The same goes for the edges. Using select_last_edges_created() will make a selection of edges that includes all edges in the graph.

# Select the last edge in the graph's EDF and confirm
# that the selection was made
graph %>%
select_last_edges_created() %>%
get_edge_df_ws()


Where these functions become useful is during the building up of a graph in multiple steps. In this next example, we will start with an empty graph (using create_graph() by itself) then add a single node with add_node(), and then select that very node with select_last_nodes_created(). This selection-in-a-pipeline approach makes it easy to set node attributes because we can use the set_node_attrs_ws() function. In this example, we're adding a new timestamp attribute and assigning a value (the system time). Because the selection is persistent, we can set more attributes in further calls to set_node_attrs_ws(). When we are done with this node selection, we call the clear_selection() function and return to a clean slate of no active selection present (it removes any node or edge selection from the graph object). The example proceeds to create another new node and adding attributes, and then a new edge and edge attributes (with set_edge_attrs_ws()). This is all facilitated with the select_last_nodes_created() and select_last_edges_created() functions.

# Create a graph, node-by-node and
graph_2 <-
create_graph() %>%
select_last_nodes_created() %>%
set_node_attrs_ws(
node_attr = "timestamp",
value = as.character(Sys.time())
) %>%
set_node_attrs_ws(
node_attr = "type",
value = "A"
) %>%
clear_selection() %>%
select_last_nodes_created() %>%
set_node_attrs_ws(
node_attr = "timestamp",
value = as.character(Sys.time())
) %>%
set_node_attrs_ws(
node_attr = "type",
value = "B"
) %>%
from = 1,
to = 2,
rel = "AB"
) %>%
select_last_edges_created() %>%
set_edge_attrs_ws(
edge_attr = "timestamp",
value = as.character(Sys.time())
) %>%
clear_selection()

# View the new graph
graph_2 %>% render_graph()


The graph shows two nodes connected together. Nothing more, nothing less. The more interesting views of the data are of the node and edge data frames, which now have several attributes set. Let's have a look at the graph's internal node data frame...

# Inspect the new graph's NDF
graph_2 %>% get_node_df()


...and let's inspect the graph's internal edge data frame.

# Inspect the new graph's EDF
graph_2 %>% get_edge_df()


As can be seen, immediately invoking select_last_nodes_created() or select_last_edges_created() after addition of new nodes or edges can be useful for working with the newly made nodes/edges. Many functions ending with _ws() operate specifically on selections of nodes or edges.

## Try the DiagrammeR package in your browser

Any scripts or data that you put into this service are public.

DiagrammeR documentation built on July 2, 2020, 3:19 a.m.