Estimation	Visualisation	Simulation	BIDS pipeline	Decoding	Statistics	MixedModelling

Package (-family) to perform linear / GAM / hierarchical / deconvolution regression on biological signals.

This kind of modelling is also known as encoding modeling, linear deconvolution, Temporal Response Functions (TRFs), linear system identification, and probably under other names. fMRI models with HRF-basis functions and pupil-dilation bases are also supported.

Getting started

🐍Python User?

We clearly recommend Julia 😉 - but Python users can use juliacall/Unfold directly from python!

Julia installation

Click to expand

The recommended way to install julia is juliaup. It allows you to, e.g., easily update Julia at a later point, but also test out alpha/beta versions etc.

TL:DR; If you dont want to read the explicit instructions, just copy the following command

Windows

AppStore -> JuliaUp, or winget install julia -s msstore in CMD

Mac & Linux

curl -fsSL https://install.julialang.org | sh in any shell

Unfold.jl installation

using Pkg
Pkg.add("Unfold")

Usage

Please check out the documentation for extensive tutorials, explanations and more!

Tipp on Docs

You can read the docs online: - or use the ?fit, ?effects julia-REPL feature. To filter docs, use e.g. ?fit(::UnfoldModel)

Here is a quick overview on what to expect.

What you need

using Unfold

events::DataFrame

# formula with or without random effects
f = @formula 0~1+condA
fLMM = @formula 0~1+condA+(1|subject) + (1|item)

# in case of [overlap-correction] we need continuous data plus per-eventtype one basisfunction (typically firbasis)
data::Array{Float64,2}
basis = firbasis(τ=(-0.3,0.5),srate=250) # for "timeexpansion" / deconvolution

# in case of [mass univariate] we need to epoch the data into trials, and a accompanying time vector
epochs::Array{Float64,3} # channel x time x epochs (n-epochs == nrows(events))
times = range(0,length=size(epochs,3),step=1/sampling_rate)

To fit any of the models, Unfold.jl offers a unified syntax:

Overlap-Correction	Mixed Modelling	julia syntax
		fit(UnfoldModel,[Any=>(f,times)),evts,data_epoch]
x		fit(UnfoldModel,[Any=>(f,basis)),evts,data]
	x	fit(UnfoldModel,[Any=>(fLMM,times)),evts,data_epoch]
x	x	fit(UnfoldModel,[Any=>(fLMM,basis)),evts,data]

Comparison to Unfold (matlab)

Click to expand

The matlab version is still maintained, but active development happens in Julia.

Feature	Unfold	unmixed (defunct)	Unfold.jl
overlap correction	x	x	x
non-linear splines	x	x	x
speed		🐌	⚡ 2-100x
GPU support			🚀
plotting tools	x		UnfoldMakie.jl
Interactive plotting			stay tuned - coming soon!
simulation tools	x		UnfoldSim.jl
BIDS support	x		alpha: UnfoldBIDS.jl)
sanity checks	x		x
tutorials	x		x
unittests	x		x
Alternative bases e.g. HRF (fMRI)			x
mix different basisfunctions			x
different timewindows per event			x
mixed models		x	x
item & subject effects		(x)	x
decoding			UnfoldDecode.jl
outlier-robust fits			many options (but slower)
🐍Python support			via juliacall

Contributions

Contributions are very welcome. These could be typos, bugreports, feature-requests, speed-optimization, new solvers, better code, better documentation.

How-to Contribute

You are very welcome to raise issues and start pull requests!

Adding Documentation

1. We recommend to write a Literate.jl document and place it in docs/literate/FOLDER/FILENAME.jl with FOLDER being HowTo, Explanation, Tutorial or Reference (recommended reading on the 4 categories).
2. Literate.jl converts the .jl file to a .md automatically and places it in docs/src/generated/FOLDER/FILENAME.md.
3. Edit make.jl with a reference to docs/src/generated/FOLDER/FILENAME.md.

Contributors

Judith Schepers 🐛 💻 📖 ✅ 🤔 ⚠️	Benedikt Ehinger 🐛 💻 📖 ✅ 🤔 ⚠️ 🚇 ⚠️ 🚧 👀 💬	René Skukies 🐛 📖 ✅ 💻 🤔	Manpa Barman 🚇	Phillip Alday 💻 🚇	Dave Kleinschmidt 📖	Saket Saurabh 🐛
suddha-bpn 🐛	Vladimir Mikheev 🐛 📖	carmenamme 📖	Maximilien Van Migem 🐛	Till Prölß 📖 🐛	Leon von Haugwitz 🐛	Jordan Deakin 🐛
Sanaz 📖

This project follows the all-contributors specification.

Contributions of any kind welcome!

Citation

For now, please cite

and/or Ehinger & Dimigen

Acknowledgements

This work was initially supported by the Center for Interdisciplinary Research, Bielefeld (ZiF) Cooperation Group "Statistical models for psychological and linguistic data".

Funded by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany´s Excellence Strategy – EXC 2075 – 390740016