i think it's valuable in making the more easily accessible.

imagine a web dev who needs facial recognition but isn't savvy enough to put it together by hand?

having a simple API solves so many problems quickly

Even if the core of the algorithm is released, there can still exist better proprietary implementations. Google Pagerank, Random Forests, ROOT/SIFT, etc are examples of 'best in field' algorithms that aren't available for free, public use.

Pagerank is just an eigenvector calculation, and for a sparse graph it's linear in |V|+|E|. It's pretty easy to implement, I don't know what kinds of improvements you'd want. There's a ton of random forest implementations out there, what's the closed one that's so much better?

Also rootSIFT is patented so good luck with that.

Google has all sorts of stuff built on top of PageRank for their search engine - it's probably safe to say that their implementation is thus 'better' than the released algorithm.

One of the RF creators consults with a company (http://www.salford-systems.com/products/randomforests) on, I suppose, better implementations. Some people have claimed it's better than the R implementation, I don't know.

RootSIFT being patented is kind of my point - the authors created a (good) algorithm and haven't released it for free use. It's within these kinds of niches that Algorithmia can succeed.

I think there's some confusion. Pagerank is the leading eigenvector of the graph (G+1x) where G is an input graph and x is a weighting term.

This is not at all the same as modern Google ranking techniques, such as Panda, Penguin, and Hummingbird.