README.md

In FvD/conceptr: Tools to work with concept hierarchies

Please look into data.tree package first if you need a way to work with hierarchical data structured as trees. The data.tree package is mature and probably does what you need.

This is a convenience package that includes amongst other things the posibility to read trees in self-referencing tables. There is no release yet, and is under active development. Be careful if you use the current development version as you may likely run into be errors.

conceptr

Concept hierarchies are common in many knowledge areas and applications of statistics. They can be deceptively simple, as they are easy to intuit, but can offer annoying problems while working with them as data set.

This package was made with a specific concept hierarchy in mind where the hierarchy is represented by a table with id, concept-name and parent-id. This structure that is often found in biological data, in competency data (Human Resource Management) and business-inventories.

A typical example could be as follows:

Level1:      A           B           C          D
             |           |           |     
Level2:    AA AB      BA BB BD      CA    
           |                         |
Level3:   AAA                     CAA CAB
                                   |
Level4:                          CABA

This kind of tree structure can be used for instance on hierarchies such as tree > Oak > White Oak. Or Technical Skills > Programming > R, or Sales > EMEA > Europe > Lithuania. These are also called multilevel classes, or multilevel features (in Machine Learning applications).

The goal of the package conceptr package is to help you when you need to work with these kinds of data by allowing you to make the hierarchy corresponding to the data explicit so that it becomes easier to:

1. Roll-up data into higher order concepts, even if they are implied but not included in the data set itself. For example, you have White Oaks and Chihuaha Oaks, but no entry for "Oaks". That higher order concept "Oak" is implicit in the data set and can be added by summing sub-level concept.
2. Visualize the hierarchies as a tree using networkD3.
3. Inspect the data to see whether it conforms with the given hierarchy
4. Visualize data in the hierarchy using ternary diagrams.

Installation

The package is still far from complete, so please be mindful of errors if you are going to use this. By all means report or fix any issues that you may find.

To install makes sure you have the package devtools installed and run:

devtools::install_github("fvd/conceptr")

Usage

Data formats

The library is still primitive and not tested for all possible exceptions. You can help by reporting any issues you find with your data so that the package can be made more generic.

At the moment a number of conventions are imposed:

1. The hierarchy data frame has three columns with names c("id","name", "parent_id")
2. The column names in the data data frame need to match the names in hierarachy$name

Note that to match the names in the hierarchy and data data frames both the hierarchy$name column and column names in data will be normalized with make.names.

Features

Aggregate data

Aggregate all data from all the sub-levels of a concept by summing the observations. So in the example aggregate_byname(D) returns D and aggregate_byname(B) returns B+BA+BB+BD. We need to reference the hierarchy to use so the full example would be:

 aggregate_byname(data, hierarchy, "Level1-D")

and

aggregate_byname(data, hierarchy, "Level1-B")

The full names in the test data base include the levels, so Level1-A, Level2-AA etc.

Visualize hierarchies

Plot selected concepts

Make exploratory analysis of the data at different levels easier by plotting by name.

visualize_ternary(data, hierarchy, c("A", "BA", "CAA"))

Planned

• Visualize hierarchy (create tree with networkD3)
• Inspect hierarchy (check the structure of the hierarchy)

Example data

There is a small example data set included that you can compare to your own. Please note that you preferably use R-style names (no spaces, no non-ASCII) for the column names in the data file. The package will make.normal any names that it will find.

To load the test data from the inst folder in the package:

data <- read.csv(system.file("data.csv", package="conceptr"))
hierarchy <- read.csv(system.file("concept-hierarchy.csv", package="conceptr"))

FvD/conceptr documentation built on May 6, 2019, 5:07 p.m.