Hello everyone, New Julia user here. Suppose I have a vector of strings, and I’d like to generate a dummy variable for each unique value that appears in that string. For instance: data = ["a", "b", "a", "a", "b"] I want to create a matrix of dummy variables along the following lines: mat_col1…

That’s very weird behavior from StatsModels. It’s not what I would have expected… Maybe @dave.f.kleinschmidt can pop in and let us know what’s going on. One thing though, is taking a look at the docs of StatsModels.ContrastsMatrix with ?, the 2nd argument is a levels, not the values themselves. So …

This is probably closer to what you want julia> df = DataFrame(x = rand(["a", "b"], 10), y = rand(10)); julia> mf = ModelFrame(@formula(y ~ 0 + x), df, contrasts = Dict(:x => DummyCoding())) |> modelmatrix 10×2 Matrix{Float64}: 0.0 1.0 0.0 1.0 1.0 0.0 1.0 0.0 0.0 1.0 0.0 1.0 0.0 1.0 …

Well, it’s not as good as previous solution, but very fun (in its own way). Inspired by Flux one-hot encoding using SparseArrays function make_dummy(data::AbstractVector) udata = unique(data) inds = map(x -> findfirst(==(x), udata), data) sparse(1:length(data), inds, ones(Int, length(da…

Hi @rubaiyat and welcome! There are lots of tools for doing this in Julia…I think MLJ has some as well, as does Flux: One-Hot Encoding · Flux (as I think someone else mentioned). If you’re just looking to do a one-off conversion, StatsModels might be a bit overkill for that. I’d recommend using i…

Yeah that’s what’s happening…we should probably throw an error StatsModels v0.6.24 will error ( https://github.com/JuliaStats/StatsModels.jl/pull/231 ) if the levels are not unique! I think this is first time I know of this coming up in practice though :slight_smile:

Also, to further clarify what ContrastsMatrix is doing/means: the contrasts matrix is essentially the embedding matrix, where each row corresponds to the model matrix row that will be generated for the corresponding level of the data. It tells you HOW to generate the one-hot (etc.) embedded represe…

Here is a pretty complete list for how to do one-hot encoding. It looks like the easiest might be StatsBase.indicatormat

@danielw2904 , which for the OP case could be slightly adjusted to: julia> permutedims(unique(data)) .== data # by @Mattriks 5×2 BitMatrix: 1 0 0 1 1 0 1 0 0 1

Generating dummy variables from a vector of strings (one-hot encoding)

New to Julia

danielw2904 July 30, 2021, 11:12pm 9

My favorite All the ways to do one-hot encoding - #4 by Mattriks

Topic		Replies	Views
All the ways to do one-hot encoding General Usage	30	11753	October 20, 2024
Learning Julia: Writing a onehot encoder Tooling	5	1534	October 23, 2019
Encoding categorical variables within a matrix Machine Learning machine-learning	3	2774	December 28, 2019
Dummy Encoding(One hot encoding) from PooledDataArray General Usage question	10	3269	June 9, 2017
Having trouble with StatsModels.jl `modelmatrix` to replicate R results Statistics	1	97	April 11, 2025

Generating dummy variables from a vector of strings (one-hot encoding)

Related topics