Lottery data is complete noise, the only structure is by definition of the game... like pick 3 has 3 digits in a combo, 0 to 9 in spots 1, 2 and 3. However, there are exactly 0 relationships to past draws and 0 relationships between the positions.

One of the things I have come to accept over the last 4 months of switching to bioinformatics as a hobby is that biological data is also noisy, but it is still structured, even if it is structured by millions of years of evolution. In many of the data projects using genomic data, there are no "Exact" matches, there could be variation in DNA or RNA that changes one nucleotide, but the overall energy available still allows a ribosome to recognize an attachment site so it can begin constructing the protein chain from the instructions.

There is much about probability involved. One of the last challenges I had was to construct a Markov Chain Monte Carlo (MCMC) simulator to allow the discovery of a regulatory motif (patterns of A, C, G or T) that has mutated over time to prevent an exact match. In 7 more chapters I will hit upon Hidden Markov Models (HMMs). So the study that is about biological computing is actually turning into a master class on probability over noisy data.

There are many ideas about applying some of this back to the lottery problem, but none actionable just yet. Just over the last week or so, I feel that I better understand the problem of getting stuck in a local minimum during the gradient descent phase of machine learning by seeing how different approaches to identifying patterns in seemingly random strings is not only possible, but fundamental to the advancement of biology.

One such area is in identifying repeated strings when you do not know in advance how long these strings will be... that seems to have a direct application in discovering patterns that are not obvious because they are not equal... but their variance is small enough to be identified!

More later, this textbook has 11 chapters and I just finished chapter 3...