PCA then ICA for taste templates

[hfgem]

• PCA on binned-fr smoothed taste responses
• Take the top n components that explain the most variance (store these for use in re-dimensionalizing all burst/potential replay events later)
• Re-dimensionalize the taste responses to the top n components
• Run ICA on the lower-dimension taste responses to pull out template components
• Look for which components align with each taste to get the separate templates and store templates
• Use templates and re-dimensionalized bursts/binned firing rates to determine points of potential replay