Fine control over which samples end up in train/validation sets

We have released a feature to control how data is split between your train/validation set fixing issues where very similar data is mixed between both train / validation set (thus giving you a false sense of high accuracy, see e.g. Does Building Model with EI’s Random Split will leads data Leakage?).

You can find the docs and some background on the feature here: Metadata - Edge Impulse Documentation

2 Likes