search

wendtke / psyphr

legacy repo for R package suite for psychophysiological data; see github.com/psyphr-dev
1 stars 1 forks source link

Evaluate BIDS Schema #52

Closed iqis closed 5 years ago

iqis commented 5 years ago

Tom Johnston on Twitter suggested "BIDS", Brain Imaging Data Structure here.

Goals:

Python API:

Validation Tool:

wendtke commented 5 years ago

See #51 also

wendtke commented 5 years ago

@iqis @MalloryJfeldman

@geanders suggested we do not try to address raw data in {psyphr}. Instead, we might think about creating a suite of related packages under a common umbrella term: one for raw data, one for output, one for complex analysis, etc.

MalloryJfeldman commented 5 years ago

Yeah to summarize my argument: We've worked for the past year @ NEU building out appropriate feature detection algorithms in python and while they've come to perform acceptably - they are only really optimal when there's really large quantities of data (we're using these algorithms to perform bulk analysis on over 5,000 hours of physio data). If our goal is to encourage best practices; then atleast for now I think we want to encourage people to place eyes on their data (and not just a select few segments, but all segments if possible). Automation via algorithms (from my perspective) is not yet sophisticated enough to capture the wide range of idiosyncrasies in physiological signal morphologies (especially when it comes to impedance, and in some regards, EDA). As of now, we have to throw out viable data that cannot be run through our algorithms, but could be successfully scored by human observers. I would LOVE to see data scoring get automated (obviously, since we are trying to contribute to that with our python code which I believe our engineering team plans to make open source) however, if Psyphr is trying to corner the "Responsible, best-practice" niche I think we want to steer clear for right now. Just my opinion though.

wendtke commented 5 years ago

This seems two-fold: 1) file directory structure and naming conventions and 2) feature detection algorithm and post-processing.

Although we might be able to address these issues in the future, I would like to keep {psyphr} focused and exclusive. Maybe one day we have a collection of packages under an umbrella package (e.g., {tidyverse}).