Purpose: a general and flexible function for doing inverted encoding models.

Needs: a flexible function that can work with 1 or more conditions, variable numbers of electrodes, different EEG signals. Assuming 1 subject at a time.

Outputs: Channel response function structure, with fields

1. Setting paths and dependencies
   • Input: actual pathnames and dependencies
   • Output: EM structure (encoding model)
     • Datapath, analysispath, sjID, dependencies
2. Set analysis parameters
   • Input: EM structure from step 1
   • Use IEM_settings (Tom’s, includes basis set, freqs of interest, etc.)
   • Number of conditions included, fixed or independent variable
   • Cross-validation type
   • Output: Exploded fieldname variables from the EM structure (this is convenience so one doesn’t have to refer to the structure.fieldname everytime).
3. Get the data
   • Input: paths and preprocessing info from exploded EM struct
   • Load the data file
   • Assuming preprocessed data stored in a particular format
   • Deal with last stages of preprocessing (e.g., remove rejected trials, deal with bad channels, etc.)
   • Number of trials for each condition and each bin of the stimulus space (include check for whether the ntrials is reasonable)
   • If needed, rearrange dimensions of the data matrix to be ntrials x nelecs x ntimepoints
   • Output: ntrials x nelecs x ntimepoints data matrix, and stuff from (e). eeg
4. Loop through frequencies and n basis sets
   • Put the data into frequency domain
     • Input: eeg from step 3, relevant info from EM structure
     • Butterworth
     • Hilbert
     • “Downsample” if desired
     • Output: eegFreqDecomp ntrials x nelecs x ntimes
   • Do the modelling
     • Input: eegFreqDecomp, basis set
     • If doing block method (e.g., Foster et al., 2016), do it and iterate
     • Structure the data for the cross-validation type
     • Convert to power (total and/or evoked)
     • Loop through cross-vals (i.e., LOO or k-fold)
     • Do IEM
       • Select training data
       • Select test data
       • Estimate weights
       • Invert on test data
       • To shift or not to shift
     • Output: centered and/or uncentered CRFs
5. Cleanup
   • Package all the analysis details & results into the EM structure
   • Save
   • Ciao