NWB Workshops and Hackathons
Back to the projects listComplex Behavior
Key Investigators
- Akshay Jaggi
- Talmo Pereira
- Arie Matsliah
- Kanishk Jain
- Jim Bohnslav
- Laurel Keyes
Project Description
Write a set of basic neurodata types to cover commonly used behavioral data in novel complex behavioral quantification schemes
Objectives
- ndx-labels: both supervised and unsupervised behavioral labels for single- and multi- animal sessions
- ndx-datatransforms: structuring matrix transformations (PCA, NMF, tSNE, etc.) and metadata
- ndx-multi: allowing multi-animal coordination of NWB sessions. Likely will be an extension of objects like Subject
Approach and Plan
Progress and Next Steps
- ndx-labels: Consolidated ndx-labels and ndx-datatransforms into one extension called ndx-labels. We wrote up a simple schematic for how to relate a behavioral video to a general label class and how to relate a general label class to a potential learned, low-dimensional “representation” of the raw data (say PCs or tSNE factors). We spec’d this schema, wrote a roundtrip test, and finished with a working example .py file. Potential extensions include making the representation class more flexible to include relevant metadata for certain representations (say PC loadings / eigenvalues), writing unit tests, testing that labels and representations could be used with neural data instead of a video, and more ideas.
- ndx-pose: Extended ndx-pose to include support for part grouping procedures for creating animal instances and support for animal identity estimation in multi-animal experiments. Wrote unit tests. Potential extensions include writing up a schema diagram for documentation, writing a round trip test, integrating this with ndx-labels, and more ideas. Special shoutout to Talmo for working on wrapping this up.
Materials
Overview Presentation: https://docs.google.com/presentation/d/1LFkAbY-lNrwpxQkO1J1S8lcBlJCF8kUQ8EJlSkTb-Z8/edit?usp=sharing