Prior Utilities

champollion.process_prior_data(prior_data)

Center and normalize prior features cell-wise.

The prior representation is row-centered, then nonzero rows are normalized to unit Euclidean norm. After this transformation, dot products between rows correspond to centered cosine similarities. Rows with zero norm are left as zeros to avoid NaNs.

Parameters:

prior_data – Array-like prior representation with cells in rows and prior features in columns.

Returns:

Processed prior representation.

Return type:

numpy.ndarray