NiFeatures is a collection of Scikit-Learn compatible transformers, tools that clean, reduce, expand or generate feature representations for machine learning models (Click here for a more detailed description of scikit-learn transformers).
Spatial patterns of neural activity can vary widely between individuals due to genetic and experiential factors, and, in neuroimaging analyses, comparability across individuals is further compromised by various technical factors (e.g. inaccurate spatial standardization). Yet, interindividual voxel-by-voxel correspondence is an implicit assumption in most type of analyses, including brain-based predictive modelling. We address this issue with our Displacement Invariant Transformer (DIT), which generates new meaningful features by relaxing this assumption.
Check out the concept behind the DIT with our IBRO2023 poster.
An hyperparameter search class tailored for our Displacement Invariant Transformer. Give a look at how it works in our example notebook (see the "Example Usage" section).
Available soon...
A simple usage example for each transformer can be found in its dedicated notebook:
Available notebooks:
- Displacement Invariant Transformer: notebooks/DIT_example.ipynb
- coming soon...
You can find the NiFeatures documentation following this link here (not available, yet).