POS sometimes contains only one element

[asanchezpuente]

If the threshold is high and (probably) the dimensional reductions is not very high.

As a consequence, numpy.tril(numpy.corrcoef(bip.data[:,POS].T), -1) fails