Architecture
Data flow
Assembly provides stories, brain data, timestamps, and split indices
Feature extractors produce time series per story (text or speech)
Downsampler aligns features to TRs
FIR expansion builds delayed feature matrices
Trainer structures data (concat or train/test split)
Model fits/predicts and metrics are logged/saved
Key components
encoding.assembly
: dataset abstractions and loadersencoding.features
: text/speech feature factories and cachesencoding.downsample
: resampling utilitiesencoding.features.FIR_expander.FIR
: delay expansionencoding.models
: ridge, nested CV, sklearn wrappersencoding.trainer.AbstractTrainer
: orchestration and loggingencoding.plotting
: plotting and experiment loggingencoding.utils
: caches, model saver, utilities