We took some accelerometer data at 75 Hz with a different device than we will use at 50 Hz or slower. So I think the data should be decimated, then retrained. But I read here that decimation is only available in enterprise. That is too bad, because it seems like a very basic feature anyone would want to try just to reduce the data and A/B test it to understand if ML would be effective at lower sample rates without collecting the data all over again.
Thank you for the feature request! Funnily enough, our engineering team is actually working on implementing a downsampling block for time series data into the Studio as we speak. I will update this post when this feature gets pushed into production.
This is now released @Panometric: https://www.edgeimpulse.com/blog/introducing-downsampling-for-time-series-data