@ItsNeuronal @karlrohe It's not a matter of being restrictive. It is what PCA is. Even if originally that was not understood. Same thing with other algorithms, for example neighbor joining: https://t.co/tHaYgZb1nq https://t.co/xe3ulkezUm
@NimwegenLab @KenjiEricLee @Bun_Without_B @dbweissman @benoitbruneau @drbstewart What I'm looking for is something analogous to this kind of theorem for neighbor-joining (there are many such theorems for all sort of computational biology algorithms): https
@JSEllenberg There are examples of heuristics based on intuition that turn out to have provably good properties. The neighbor-joining algorithm in phylogenetics comes to mind (see, e.g. https://t.co/xe3uljWqGe, although even NJ is outperformed by other met
@akshaykagrawal @KeithComplexity Thanks. But this is not a theorem. It's intuition by analogy. Sometimes heuristics are developed before theory for them is understood, and it's interesting to reverse engineer understanding. I've done such work myself, see
RT @lpachter: One consequence of this was that I ended up reflecting on how well I understood NJ. I could explain the steps but realized I…
RT @lpachter: One consequence of this was that I ended up reflecting on how well I understood NJ. I could explain the steps but realized I…
RT @lpachter: One consequence of this was that I ended up reflecting on how well I understood NJ. I could explain the steps but realized I…
One consequence of this was that I ended up reflecting on how well I understood NJ. I could explain the steps but realized I didn't really understand why it worked when it did. Ended up writing several papers on the subject: https://t.co/xe3uljWqGe
@jakeberv makes an important point. NJ is a consistent estimator of phylogeny https://t.co/WIpcdwBw9t but that is only if the distances used as input are correct