lib5c.contrib.iced.balancing module

Module for interfacing with the external iced Python package, which provides access to the ICED matrix balancing algorithm.

lib5c.contrib.iced.balancing.iced_balance_matrix(matrix, max_iter=3000, eps=0.0001, norm='l1', imputation_size=0)[source]

Convenience function wrapping the ICE_normalization function from the external iced Python package, which balances a counts matrix using the ICE algorithm.

Parameters:
  • matrix (np.ndarray) – The counts matrix to balance.
  • max_iter (int) – The maxiumum number of iterations to try.
  • eps (float) – The relative size of error before declaring convergence.
  • norm ({'l1', 'l2'}) – What norm to use as a distance measure.
  • imputation_size (int) – Pass an int greater than 0 to replace NaN’s in the matrix with a local median approximation. Pass 0 to skip imputation.
Returns:

The first element of the tuple is the balanced matrix. The second element is the bias vector.
Return type:

Tuple[np.ndarray, np.ndarray]