Why Julia for EEG Research?
Most EEG analysis today happens in MATLAB (EEGLAB, FieldTrip, ERPLAB) or Python (MNE-Python). These are well-established packages with many years of development, large user communities, and broad feature sets — they will typically offer more functionality than EegFun.jl. Each ecosystem has clear strengths, and researchers should choose the tool that best fits their needs. This page explains why EegFun.jl is built on Julia and where the trade-offs lie.
Strengths
Fully open source — language and libraries. Julia itself and its entire package ecosystem are open source. MATLAB-based toolboxes such as EEGLAB, FieldTrip, and ERPLAB are open source, but they still require a proprietary MATLAB licence to run. GNU Octave provides a free alternative, but compatibility with these toolboxes is not 100%, and performance and graphics support can differ significantly.
One language from prototype to production. Scientific computing often hits a "two-language problem": researchers prototype in a high-level language and then rewrite performance-critical code in C or Fortran. In practice this means R code calling Rcpp/C++, Python relying on C extensions (NumPy, SciPy, MNE's compiled backends), and MATLAB using MEX files. Julia is designed to avoid this split — user-level code compiles to efficient native code, so there is no need to drop into a second language for speed. In EegFun.jl, there is no hidden C or Fortran layer; the analysis code you read is the code that runs.
Performance without boilerplate. Julia's JIT compiler generates machine code specialised to the types you actually use. For numerically intensive EEG workflows — filtering, time- frequency decomposition, permutation statistics — this often matches C/Fortran performance.
Multiple dispatch for clean extensibility. Julia's type system and multiple dispatch make it straightforward to extend existing functions to new data types. In EegFun.jl this means, for example, the same lowpass_filter! function works on continuous data, epoched data, and ERP averages.
Built-in package manager and reproducibility. Every Julia project records its exact dependency versions in Manifest.toml. Collaborators can reproduce your environment with a single Pkg.instantiate() call, without conda environments or Docker containers.
Trade-offs
Time to first execution (TTFX). Julia compiles code the first time it runs in a session, which means the first call to a function can take noticeably longer than subsequent calls. Python users familiar with Numba will recognise this pattern — both compile code to fast machine code on first use — but in Julia, the compilation applies to the entire language rather than individual decorated functions. This "compilation latency" has improved substantially in recent Julia versions (1.9+ introduced package images, and 1.12 brings further improvements), but it remains more noticeable than starting a Python or R session. In practice, most users start a Julia session once and keep it running.
Smaller ecosystem. The Julia ecosystem for EEG is young compared to MATLAB and Python. Packages like EEGLAB (2004), FieldTrip (2011), and MNE-Python (2014) have over a decade of development, thousands of users, and extensive functionality that EegFun.jl does not yet match.
Tooling is still maturing. IDE support, debugging, and profiling tools exist (VS Code with the Julia extension is quite capable), but they are not as mature or polished as the equivalent MATLAB IDE or established Python tool-chains (PyCharm, Jupyter ecosystem). The debugger, in particular, can be slow on large codebases.
Package loading time. Large packages with many dependencies take time to load (using EegFun is not instant). The very first using EegFun after installation is particularly slow because Julia precompiles (partly) the package and all of its dependencies — this is a one-off cost. Subsequent using EegFun calls in new sessions are considerably faster, but still slower than import mne in Python.
If you are coming from MATLAB or Python, the [MATLAB-Python-Julia cheat sheet](https://cheatsheets.quantecon.org/) is a helpful reference for translating familiar idioms.