NeuroPype 2024
We're excited to announce our next major release of the NeuroPype signal processing suite! This release brings powerful new Bayesian modeling capabilities, automated sleep scoring, and a host of engine, API, and usability improvements.
A new Bayesian package
The headline addition is a brand-new Bayesian package with nearly 20 nodes that let you build fully Bayesian models for machine-learning and deep-learning classifiers — for example, state decoding from biosignals — entirely within a NeuroPype pipeline. It includes a complete toolkit for probabilistic inference: MCMC samplers (NUTS, HMC, mixed and discrete variants), Stochastic Variational Inference, mean-field and multivariate-normal approximations, distributions, and the building blocks for expressing rich hierarchical models as a graph.
Automated sleep scoring & expanded ML/DL
This release also introduces automated sleep scoring: the new HypnogramHMM node computes a hypnogram directly from EEG using a hidden Markov model. We've expanded our machine learning and deep learning methods too, with probability calibration (Platt scaling), ensemble predictors, trial-aggregate predictors, and a robust (Huber) loss option for Hierarchical Discriminant Component Analysis.
Engine optimizations
Under the hood, the engine has been optimized for both performance and memory usage, with support for multiple LAP solver backends, more robust handling of edge cases, and new modes for running in low-memory environments. You can now also define "global" settings that can be wired in anywhere in a pipeline — very handy for keeping large pipelines manageable.
Easier pipeline control over the API
Controlling pipelines over the NeuroPype API is easier than ever, with new endpoints to play, pause, resume, stop, and configure a running pipeline, plus simpler endpoints to load and run pipelines. (All the old endpoints still work, of course.)
Visualization & usability
The TopoPlot viewer can now generate figures from offline data and render interpolation and contour lines (with thanks to Lachlan Barnes at the University of Auckland for contributing the contours option), and the NeuroPype Control Panel is friendlier to use with new file pickers for selecting pipelines and files — plus, as always, plenty of improvements, node documentation clarifications, and bug fixes throughout.
Grab the latest release from your NeuroPype account, or sign up today if you're not yet a NeuroPype user. We greatly appreciate any feedback you may have, especially from those using the Academic Edition.