MLJTransforms.jl

A Julia package providing a wide range of categorical encoders and transformers to be used with the MLJ package. Transformers help convert raw features into a representation that's better suited for downstream models. Meanwhile, categorical encoders are a type of transformer that specifically encodes categorical features into numerical forms.

Installation

import Pkg
Pkg.activate("my_environment", shared=true)
Pkg.add(["MLJ", "MLJTransforms"])

Quick Start

For the following demo, you will need to additionally run Pkg.add("RDatasets").

using MLJ
import RDatasets

# 1. Load Data
X = RDatasets.dataset("HSAUR", "Forbes2000");

# 2. Load the model
encoder = FrequencyEncoder(
    features=[:Country, :Category],     # The categorical columns to select
    ignore=false,                       # Whether to exclude or include selected columns
    ordered_factor = false,             # Whether to also encode columns of ordered factor elements
    normalize=true                      # Whether to normalize the frequencies used for encoding
    )


# 3. Wrap it in a machine and fit
mach = fit!(machine(encoder, X))
Xnew = transform(mach, X)

Available Transformers

See complete list of transformers in this package.

In MLJTransforms we denote transformers that can operate on columns with Continuous and/or Count scientific types as numerical transformers. Meanwhile, categorical transformers operate on Multiclass and/or OrderedFactor scientific types. Most categorical transformers in this package operate by converting categorical values into numerical values or vectors, and are therefore considered categorical encoders. We categorize categorical encoders as follows:

Some transformers in this package can even operate on both Finite and Infinite scientific types or other special scientific types (eg, to represent time). To learn more about scientific types see the official documentation.