Closed treo closed 6 years ago
Thank you very much, Paul. I'll include this in the next round. I'll have to decline the PR since it slightly breaks the rest of the code, but I'll make sure I'll include it manually and it will be there in the next commit.
I'm glad that an explanation was found. I hope Dragan updates the post. But that does not undermine his larger point - that his library is easier to use, and does the right thing under the hood.
A response to this: https://dragan.rocks/articles/18/Neanderthal-vs-ND4J-vol3 Also, this change was added in the last commit: https://github.com/uncomplicate/neanderthal/commit/e09bedeec4243f0090356c311389665249d73a6a
Your original version computes
M = A(B(C(D(EF))))
. This also addsM = ((((AB)C)D)E)F
. Depending on the actual matrix sizes one is a lot faster than the other.