mescla.prep.missing.largest_complete_block¶

mescla.prep.missing.largest_complete_block(table, min_species=2)[source]¶

Find the complete sub-matrix retaining the most data.

Greedily drops the species that buys the most, stopping when dropping another would retain less data overall. The objective is complete_samples x species_kept, not the sample count alone: a block of 58 samples with 2 species is worse than 51 samples with 5, and optimising sample count alone walks straight into the former.

Not guaranteed optimal – the exact problem is combinatorial – but it is fast and, in practice, hard to beat by hand.

Parameters:

min_species (int, default 2) – Refuse to go below this many species. ne end-members need ne - 1.
table (Any)

Returns:

ChemTable – A complete block: no missing values.

Return type:

ChemTable

Examples

>>> block = largest_complete_block(ragged)
>>> block.shape
(51, 5)