pkgnet : Network Analysis of R Packages

Description

R packages can be complex bodies of code and functionality with hard-to-intuit underlying structure. pkgnet provides tools to analyze and understand that structure through the lens of network theory.

Package Report

The simplest way of using pkgnet is through the function CreatePackageReport, e.g.,

    CreatePackageReport("lubridate")

This will create a standalone HTML report containing analyses of various aspects of the specified subject package.

For more info, check out our introductory vignette with

    vignette("pkgnet-intro")

Individual Reporters

Reporters are the basic modules of functionality within pkgnet. Each type of reporter is used to analyze a particular aspect of the subject package. The currently available reporters are:

DependencyReporter

analyze the recursive network of packages that the subject package depends on.

FunctionReporter

analyze the network of interdependencies of the functions defined in the subject package

InheritanceReporter

analyze the class inheritance trees for subject packages that use object-oriented programming.

SummaryReporter

get an overview of the subject package through its DESCRIPTION file.

CreatePackageReport uses a standard set of reporters by default. You can customize the reporters you want by passing in your own list of instantiated reporters, e.g.

      CreatePackageReport(
          "lubridate",
          pkg_reporters = list(FunctionReporter$new(), InheritanceReporter$new())
      )
  

You can also use reporters interactively. Once you have a reporter instantiated, check our that reporter's documentation to see what you can do.

      reporter <- DependencyReporter$new()
      reporter$set_package("lubridate")
  

Author(s)

Maintainer: Brian Burns brian.burns.opensource@gmail.com

Authors:

• James Lamb jaylamb20@gmail.com

• Jay Qi jayqi.opensource@gmail.com

See Also

Useful links:

• https://github.com/uptake/pkgnet

• https://uptake.github.io/pkgnet/

• Report bugs at https://github.com/uptake/pkgnet/issues

Base class for Graphs

Description

pkgnet uses R6 classes to define and encapsulate the graph models for representing package networks. These classes implement different types of graphs and functionality to calculate their respective graph theory measures. The base class AbstractGraph defines the standard interfaces and functionality.

Currently the only implemented type of graph is DirectedGraph.

Active bindings

Methods

Public methods

• AbstractGraph$new()

• AbstractGraph$node_measures()

• AbstractGraph$graph_measures()

• AbstractGraph$print()

• AbstractGraph$clone()

Method new()

Instantiate new object of the class.

Usage

AbstractGraph$new(nodes, edges)

Arguments

Returns

Self, invisibly.

Method node_measures()

Return specified node-level measures, calculating if necessary. See Node Measures section in DirectedGraphMeasures for details about each measure.

Usage

AbstractGraph$node_measures(measures = NULL)

Arguments

Returns

a data.table with specified node meaures as columns

Method graph_measures()

Return specified graph-level measures, calculating if necessary. See Graph Measures section in DirectedGraphMeasures for details about each measure.

Usage

AbstractGraph$graph_measures(measures = NULL)

Arguments

Returns

list with specified graph measures.

Method print()

print igraph object

Usage

AbstractGraph$print()

Returns

Self, invisibly.

Method clone()

The objects of this class are cloneable with this method.

Usage

AbstractGraph$clone(deep = FALSE)

Arguments

Abstract Network Reporter Class

Description

pkgnet defines several package reporter R6 classes that model a particular network aspect of a package as a graph. These network reporter classes are extended from AbstractGraphReporter, which itself extends the AbstractPackageReporter with graph-modeling-related functionality.

This article describes the additional fields added by the AbstractGraphReporter class definition.

Super class

pkgnet::AbstractPackageReporter -> AbstractGraphReporter

Active bindings

Methods

Public methods

• AbstractGraphReporter$calculate_default_measures()

• AbstractGraphReporter$get_summary_view()

• AbstractGraphReporter$clone()

Method calculate_default_measures()

Calculates the default node and network measures for this reporter.

Usage

AbstractGraphReporter$calculate_default_measures()

Returns

Self, invisibly.

Method get_summary_view()

Creates a summary table formatted for display.

Usage

AbstractGraphReporter$get_summary_view()

Returns

A DT object

Method clone()

The objects of this class are cloneable with this method.

Usage

AbstractGraphReporter$clone(deep = FALSE)

Arguments

Abstract Package Reporter

Description

pkgnet defines several package reporter R6 classes that analyze some particular aspect of a package. These reporters share common functionality and interfaces defined by a base reporter class AbstractPackageReporter.

Active bindings

Methods

Public methods

• AbstractPackageReporter$set_package()

• AbstractPackageReporter$get_summary_view()

• AbstractPackageReporter$clone()

Method set_package()

Set the package that the reporter will analyze. This can only be done once for a given instance of a reporter. Instantiate a new copy of the reporter if you need to analyze a different package.

Usage

AbstractPackageReporter$set_package(pkg_name, pkg_path = NULL)

Arguments

Returns

Self, invisibly.

Method get_summary_view()

Returns an htmlwidget object that summarizes the analysis of the reporter. Used when creating a package report.

Usage

AbstractPackageReporter$get_summary_view()

Returns

Self, invisibly.

Method clone()

The objects of this class are cloneable with this method.

Usage

AbstractPackageReporter$clone(deep = FALSE)

Arguments

pkgnet Analysis Report for an R package

Description

Create a standalone HTML report about a package and its networks.

Usage

CreatePackageReport(
  pkg_name,
  pkg_reporters = DefaultReporters(),
  pkg_path = NULL,
  report_path = tempfile(pattern = pkg_name, fileext = ".html")
)

Arguments

Value

an instantiated PackageReport object

pkgnet Report as Vignette

Description

Create pkgnet package report as an R Markdown vignette. This vignette can be rendered into a standard HTML vignette with the knitr::rmarkdown vignette engine into HTML vignettes upon package building. It is also compatible with #' pkgdown sites. See the vignette "Publishing Your pkgnet Package Report" for details about how to use this function, as well as our example for pkgnet.

Usage

CreatePackageVignette(
  pkg = ".",
  pkg_reporters = list(DependencyReporter$new(), FunctionReporter$new()),
  vignette_path = file.path(pkg, "vignettes", "pkgnet-report.Rmd")
)

Arguments

Default Reporters

Description

Instantiates a list of default reporters to feed into CreatePackageReport.

Usage

DefaultReporters()

Details

Default reporters are:

• SummaryReporter

• DependencyReporter

• FunctionReporter

Note, InheritanceReporter is not included in the default list.

If desired, append a new instance of InheritanceReporter to the DefaultReporters list.

ex: c(DefaultReporters(), InheritanceReporter$new())

Value

list of instantiated reporter objects

Recursive Package Dependency Reporter

Description

This reporter looks at the recursive network of its dependencies on other packages. This allows a developer to understand how individual dependencies might lead to a much larger set of dependencies, potentially informing decisions on including or removing them.

Super classes

pkgnet::AbstractPackageReporter -> pkgnet::AbstractGraphReporter -> DependencyReporter

Active bindings

Methods

Public methods

• DependencyReporter$new()

• DependencyReporter$clone()

Method new()

Initialize an instance of the reporter.

Usage

DependencyReporter$new(
  dep_types = c("Imports", "Depends", "LinkingTo"),
  installed = TRUE
)

Arguments

Returns

Self, invisibly.

Examples

\donttest{

# Instantiate an object
reporter <- DependencyReporter$new()

# Seed it with a package
reporter$set_package("ggplot2")

}

Method clone()

The objects of this class are cloneable with this method.

Usage

DependencyReporter$clone(deep = FALSE)

Arguments

See Also

Other Network Reporters: FunctionReporter, InheritanceReporter

Other Package Reporters: FunctionReporter, InheritanceReporter, SummaryReporter

Examples

## ------------------------------------------------
## Method `DependencyReporter$new`
## ------------------------------------------------



# Instantiate an object
reporter <- DependencyReporter$new()

# Seed it with a package
reporter$set_package("ggplot2")

Directed Graph Network Model

Description

R6 class defining a directed graph model for representing a network, including methods to calculate various measures from graph theory. The igraph package is used as a backend for calculations.

This class isn't intended to be initialized directly; instead, network reporter objects will initialize it as its pkg_graph field. If you have a network reporter named reporter, then you access this object's public interface through pkg_graph—for example,

reporter$pkg_graph$node_measures('hubScore')

Super class

pkgnet::AbstractGraph -> DirectedGraph

Active bindings

Methods

Public methods

• DirectedGraph$clone()

Method clone()

The objects of this class are cloneable with this method.

Usage

DirectedGraph$clone(deep = FALSE)

Arguments

See Also

DirectedGraphMeasures

Measures for Directed Graph Class

Description

Descriptions for all available node and graph measures for networks modeled by DirectedGraph.

Node Measures

outDegree

outdegree, the number of outward edges (tail ends). Calculated by igraph::degree. [Wikipedia]

inDegree

indegree, number of inward edges (head ends). Calculated by igraph::degree. [Wikipedia]

outCloseness

closeness centrality (out), a measure of path lengths to other nodes along edge directions. Calculated by igraph::closeness. [Wikipedia]

inCloseness

closeness centrality (in), a measure of path lengths to other nodes in reverse of edge directions. Calculated by igraph::closeness. [Wikipedia]

numRecursiveDeps

number recursive dependencies, i.e., count of all nodes reachable by following edges out from this node. Calculated by igraph::neighborhood.size. [Wikipedia]

numRecursiveRevDeps

number of recursive reverse dependencies (dependents), i.e., count all nodes reachable by following edges into this node in reverse direction. Calculated by igraph::neighborhood.size. [Wikipedia]

betweenness

betweenness centrality, a measure of the number of shortest paths in graph passing through this node Calculated by igraph::betweenness. [Wikipedia]

pageRank

Google PageRank. Calculated by igraph::page_rank. [Wikipedia]

hubScore

hub score from Hyperlink-Induced Topic Search (HITS) algorithm. Calculated by igraph::hub_score. [Wikipedia]

authorityScore

authority score from Hyperlink-Induced Topic Search (HITS) algorithm. Calculated by igraph::authority_score. [Wikipedia]

Graph Measures

graphOutDegree

graph freeman centralization for outdegree. A measure of the most central node by outdegree in relation to all other nodes. Calculated by igraph::centralize. [Wikipedia]

graphInDegree

graph Freeman centralization for indegree. A measure of the most central node by indegree in relation to all other nodes. Calculated by igraph::centralize. [Wikipedia]

graphOutClosness

graph Freeman centralization for out-closeness. A measure of the most central node by out-closeness in relation to all other nodes. Calculated by igraph::centralize. [Wikipedia]

graphInCloseness

graph Freeman centralization for outdegree. A measure of the most central node by outdegree in relation to all other nodes. Calculated by igraph::centralize. [Wikipedia]

graphBetweennness

graph Freeman centralization for betweenness A measure of the most central node by betweenness in relation to all other nodes. Calculated by igraph::centralize. [Wikipedia]

Function Interdependency Reporter

Description

This reporter looks at the network of interdependencies of its defined functions. Measures of centrality from graph theory can indicate which function is most important to a package. Combined with unit test coverage information—also provided by this reporter— it can be used as a powerful tool to prioritize test writing.

Details

R6 Method Support:

R6 classes are supported, with their methods treated as functions by the reporter.

• R6 methods will be named like <classname>$<methodtype>$<methodname>, e.g., FunctionReporter$private_methods$extract_nodes.

• Note that the class name used will be the name of the generator object in the package's namespace.

• The classname attribute of the class is not used. In general, it is not required to be defined or the same as the generator object name. This attribute is used primarily for S3 dispatch.

Known Limitations:

• Using non-standard evaluation to refer to things (e.g, dataframe column names) that have the same name as a function will trick FunctionReporter into thinking the function was called. This can be avoided if you don't use reuse function names for other purposes.

• Functions stored as list items and not assigned to the package namespace will be invisible to FunctionReporter.

• Calls to methods of instantiated R6 or reference objects will not be recognized. We don't have a reliable way of identifying instantiated objects, or identifying their class.

• Reference class methods are not yet supported. They will not be identified as nodes by FunctionReporter.

Super classes

pkgnet::AbstractPackageReporter -> pkgnet::AbstractGraphReporter -> FunctionReporter

Active bindings

Methods

Public methods

• FunctionReporter$calculate_default_measures()

• FunctionReporter$clone()

Method calculate_default_measures()

Calculates the default node and network measures for this reporter.

Usage

FunctionReporter$calculate_default_measures()

Returns

Self, invisibly.

Method clone()

The objects of this class are cloneable with this method.

Usage

FunctionReporter$clone(deep = FALSE)

Arguments

See Also

Other Network Reporters: DependencyReporter, InheritanceReporter

Other Package Reporters: DependencyReporter, InheritanceReporter, SummaryReporter

Class Inheritance Reporter

Description

This reporter takes a package and traces the class inheritance structure. Currently the following object-oriented systems are supported:

• S4 Classes

• Reference Classes (sometimes informally called "R5")

• R6 Classes

S3 classes are not supported, as their inheritance is defined on an ad hoc basis per object and not formally by class definitions.

Details

Note the following details about class naming:

• Reference Classes : The name passed as Class in setRefClass is used as the node name by this reporter. This is the class name that is used when specifying inheritance. The generator object returned by setRefClass does not have to be assigned and can have a different name.

• R6 Classes : The name of the generator object in the package namespace is used as the node name by this reporter. The generator object returned by R6::R6Class is what is used when specifying inheritance. The name passed as classname passed to R6::R6Class can be a different name or even NULL.

For more info about R's built-in object-oriented systems, check out the relevant chapter in Hadley Wickham's Advanced R. For more info about R6, check out their docs website or the chapter in Advanced R's second edition.

Super classes

pkgnet::AbstractPackageReporter -> pkgnet::AbstractGraphReporter -> InheritanceReporter

Active bindings

Methods

Public methods

• InheritanceReporter$clone()

Method clone()

The objects of this class are cloneable with this method.

Usage

InheritanceReporter$clone(deep = FALSE)

Arguments

See Also

Other Network Reporters: DependencyReporter, FunctionReporter

Other Package Reporters: DependencyReporter, FunctionReporter, SummaryReporter

R6 Class Representing an R Package Report

Description

pkgnet compiles one or more package reporters into a package report for a specified package. PackageReport is an R6 class that holds all of those reporters and has a method render_report() to generate an HTML report file. You can access each individual reporter and modify it using its methods if you wish.

The function CreatePackageReport() is a shortcut for both generating a PackageReport object with instantiated reporters and creating the HTML report in one call.

Value

Self, invisibly.

Active bindings

Methods

Public methods

• PackageReport$new()

• PackageReport$add_reporter()

• PackageReport$render_report()

• PackageReport$clone()

Method new()

Initialize an instance of a package report object.

Usage

PackageReport$new(
  pkg_name,
  pkg_path = NULL,
  report_path = tempfile(pattern = pkg_name, fileext = ".html")
)

Arguments

Returns

Instantiated package report object.

Method add_reporter()

Add a reporter to the package report.

Usage

PackageReport$add_reporter(reporter)

Arguments

Returns

Self, invisibly

Method render_report()

Render html pkgnet package report.

Usage

PackageReport$render_report()

Method clone()

The objects of this class are cloneable with this method.

Usage

PackageReport$clone(deep = FALSE)

Arguments

Package Summary Reporter

Description

This reporter provides a high-level overview of a package via its package DESCRIPTION file.

Super class

pkgnet::AbstractPackageReporter -> SummaryReporter

Active bindings

Methods

Public methods

• SummaryReporter$get_summary_view()

• SummaryReporter$clone()

Method get_summary_view()

Returns an htmlwidget object that summarizes the analysis of the reporter. Used when creating a package report.

Usage

SummaryReporter$get_summary_view()

Returns

Self, invisibly.

Method clone()

The objects of this class are cloneable with this method.

Usage

SummaryReporter$clone(deep = FALSE)

Arguments

See Also

Other Package Reporters: DependencyReporter, FunctionReporter, InheritanceReporter