Python Programming for Biology: Bioinformatics and Beyond

Unsupervised machine learning

Unsupervised learning is where we have an algorithm that isn’t shown a known data set

for  training.  Instead  the  algorithm  proceeds  and  optimises  of  its  own  accord  without

influence  from  a  human-imposed  set  of  standards.  Usually  the  objective  is  for  the

algorithm to organise or arrange some input data according to relatively simple rules, but

where there is nevertheless no straightforward answer. Where the data has been arranged

so  that  similar  inputs  have  been  placed  near  to  one  another  the  end  result  can  be  a

classification. Unsupervised methods are often used as a means of dimensional reduction,

where  the  input  data  has  many  independent  qualities  (dimensions)  but  you  wish  to

represent the data in only a few dimensions, say as a two-dimensional map. A biological

example  might  be  if  you  have  some  medical  data  where  you  have  an  abundance  of

different test measurements (lots of dimensions) but wish to categorise patients into a few

discrete groups, each of which will have a different treatment regime; patients within each

group will have similar test results, even though the results are multi-factorial and it is not

easy for a human to derive the groupings.