Ctrl+K

Neo 0.13.0 release notes#

2nd February 2024

This release of Neo sees a lot of changes and improvements, with a documentation rewrite, final tweaks and refinements to the object model (in anticipation of a 1.0 release in the next few months), a new approach to filtering objects of interest from complex datasets, and a large number of bug fixes and performance improvements in IO modules. 23 people contributed to this release, which is a new record for Neo!

See all pull requests included in this release and the list of closed issues.

Refinements to the Neo object model#

In order to simplify the structure of Neo datasets, and make a clearer distinction between the Block-Segment-data hierarchy and the more flexible data-grouping functionality, we no longer allow Groups to contain Segments.

We have made the handling of RegionOfInterest subclasses consistent with ChannelView, i.e., just as a ChannelView gives a view of a subset of the channels in a multi-channel AnalogSignal, so the RegionOfInterest` subclasses give views of a subset of the pixels in an ImageSequence.

Neo now has automatic handling of relationships between objects. In previous versions, the child-parent relationships between objects had to be handled manually, for example when you added a SpikeTrain to a Segment, this created a parent-child relationship, but the reverse child-parent relationship was not automatically created, leading to potential inconsistencies. Now these relationships are created automatically, as the various child lists (Segment.spiketrains, Segment.analogsignals, Block.segments, etc.) are now represented by pseudo-list objects which take care of the book-keeping behind the scenes.

As a convenience, all Neo containers (Block, Segment, Group) now have an add() method that automatically puts the object in the correct list, for example:

segment.add(signal1, event2, spiketrain3)

instead of:

segment.analogsignals.append(signal1)
segment.events.append(event2)
segment.spiketrains.append(spiketrain3)

Documentation rewrite and new theme#

The Neo documentation has had a complete overhaul, with a rewrite aimed at providing different entry points to the documentation for different user needs, using the Sphinx IPython extension to ensure code snippets are always up-to-date, and a switch to the PyData Sphinx theme.

Improved filtering#

For complex datasets, it is often necessary to extract subsets of the data from a Neo object tree by filtering based on object types, names, and annotations. Neo’s filtering capabilities have been improved, with the addition of a new filters module.

For example, suppose we have a dataset from an animal performing a behavioural task. Our objective is to retain only the trials where the animal performed correctly, for one of the four trial types in the experiment, and where the recordings met certain quality criteria. Based on an annotated Neo dataset, we can extract the spike trains of interest in a single step, without for loops or complex list comprehensions, as follows:

import neo.core.filters as nf

# ... load data, `trial` is a Segment object ...

sua_spiketrains = trial.filter(
    objects=neo.SpikeTrain,
    targdict=[
        {"sua": True},                                # only single-unit activity
        {"electrode_reject_HFC": False},              # } exclude different types
        {"electrode_reject_LFC": False},              # } of artefacts
        {"electrode_reject_IFC": False},              # } or markers of poor-quality
        {"noise": False},                             # } recordings
        {"spike_count": nf.FilterGreaterThan(10000)}  # only use neurons with a sufficiently high firing rate
    ]
)

New IO modules#

Neo now supports MED format, with the new MedIO and MedRawIO classes.

Bug fixes and improvements in IO modules#

Bug fixes and/or improvements have been made to AsciiSignalIO, BCI2000IO, BiocamIO, BlackrockIO, IgorIO, IntanIO, KlustaKwikIO, MaxwellIO, MEArecIO, NeoMatlabIO, NeuralynxIO, NixIO, NWBIO, OpenEphysIO, PlexonIO, Plexon2IO, SpikeGLXIO, and TdtIO.

Other changes#

  • RawIO classes have a method parse_header() that extracts all the information from the formats and therefore is usually a computationally heavy process. A flag is_header_parsed has now been added, to avoid re-extracting the header information if it is already available. This can produce significant speed ups.

  • Type hints have been added to the baseio, baserawio, and exampleio modules, with the aim of helping developers who wish to add a new IO module.

  • All of the code is now formatted with black.

  • The continuous integration (CI) pipeline has been sped up.

Updated dependencies#

Neo now requires NumPy version >=1.20.3.

Acknowledgements#

Thanks to Julia Sprenger, Andrew Davison, Zach McKenzie, Alessio Buccino, Moritz Alexander Kern, Samuel Garcia, Heberto Mayorquin, Joscha Schmiedt, Daniel P. Crepeau, Divyansh Gupta, Nate Dolensek, Peter N. Steinmetz, Philipp Hornauer, Robert Wolff, Jules Lebert, Kyle Johnsen, Ben Dichter, Benjamin Heasly, Baptiste Grimaud, Cody Baker, Fernando J. Chaure, @Filipe, and Matthias Klumpp for their contributions to this release.

On this page
Show Source