currently, the function that does the heavy lifting in the regression approach is hardcoded to use pixel counts, but that precludes the ability to specify more complex models (e.g. log-transforming the pixel counts, adding interactions, or including more input data).
instead, we should allow the function to take a patsy-style model formulation as a string instead of assuming we know what the user wants in this line
currently, the function that does the heavy lifting in the regression approach is hardcoded to use pixel counts, but that precludes the ability to specify more complex models (e.g. log-transforming the pixel counts, adding interactions, or including more input data).
instead, we should allow the function to take a patsy-style model formulation as a string instead of assuming we know what the user wants in this line