Possible large memory usage fix

[dkoslicki]

In the formation of the hypothetical organism portion of the matrix, a hypothetical column is added for each column of A, and for each threshold. Since the matrix A is reduced using the cutoff here, we only really need to make the hypothetical columns for the reduced A.

For large training databases and diverse samples, this will significantly reduce the memory usage (as it now briefly spikes before the basis is reduced).