## TableDistances.jl

Distances between heterogeneous tabular data
Author JuliaML
Popularity
17 Stars
Updated Last
1 Year Ago
Started In
October 2021

# TableDistances.jl

This package provides methods for computing distances between rows of general Tables.jl tables using the ecosystem of scientific types available in ScientificTypes.jl. It follows the Distances.jl interface as much as possible.

## Rationale

A common task in statistics and machine learning consists of computing distances between observations for different purposes (e.g. clustering, kernel methods). When the data is homogeneous, i.e. all the attributes have the same scientific type, one can use packages such as Distances.jl directly on the result of `Tables.matrix(table)`. On the other hand, when the table is heterogeneous, one must combine different distances for the various attributes using some weighting scheme.

## Installation

Get the latest stable release with Julia's package manager:

`] add TableDistances`

## Usage

We follow the Distances.jl interface as much as possible:

```using TableDistances
using ScientificTypes

# create an heterogeneous table
table = (a=1:3, b=rand(3), c=["A", "B", "C"], d=[1, 2, 4])
(a = 1:3, b = [0.7596581938450753, 0.6952806574889876, 0.6669145844749085], c = ["A", "B", "C"], d = [1, 2, 4])

t = coerce(table, :a => Count, :b => Continuous, :c => Multiclass, :d => OrderedFactor)
(a = 1:3, b = [0.7596581938450753, 0.6952806574889876, 0.6669145844749085], c = CategoricalArrays.CategoricalValue{String, UInt32}["A", "B", "C"], d = CategoricalArrays.CategoricalValue{Int64, UInt32}[1, 2, 4])

# compute the pairwise distance between rows
D = pairwise(TableDistance(), t)
3×3 Matrix{Float64}:
0.0      1.09707   1.75
1.09707  0.0       0.902927
1.75     0.902927  0.0```

Default distances from various packages such as StringDistances.jl are automatically chosen depending on the table schema, and weights can be specified for each attribute.

## Contributing

Contributions are very welcome. Please open an issue if you have questions.

### Required Packages

View all packages