Rebel wrote: ↑Thu Oct 01, 2020 9:38 am
Minic uses the nn-97f742aaefcd net from the SF NNUE page dated August 3 long before the SF12 release thus most likely a Sergio net which they renamed from bin to nnue. Which would indicate the Minic results are just an accident.
Accident or not, I will do 2 extra tests, 1) run Minic with another net, 2) run Rubichess with the Minic net and see if the pattern remains.
Btw, I suggest to add the release dates (at least MMDD) to the numeration of SV nets in SimEx tables, in order to sort them chronologically, otherwise it's harder to analyze temporal patterns.
What surprises me the most is the eval similarity between SF12 and Lc0, which suggests that Sergio's method of NNUE training ostensibly distills knowledge so well that SF NNUE is a marriage of a fast AB searcher with most of the knowledge captured by Leela T60 nets.
I think there is a relationship but I also think strength plays an important role, SF12 and Lc0 at one second not unlikely perform already at 3300 (or so) elo and thus engines at that level start to agree.
More technical about the relationship between NN and NNUE: the SRL list is based on deep Lc0 analysis and I don't believe the giant elo jumps NNUE nets have made are real elo points because of that relationship between NN and NNUE.
90% of coding is debugging, the other 10% is writing bugs.
Rebel wrote: ↑Thu Oct 01, 2020 10:59 pm
Regarding net-name-calling, the Sergio net numbering was fine, now the gibberish SF numbering probably is going to be the new standard, I suggest the Igel way, for Minic -> Minic-248.nnue
Something very important to me is that there is no "official net" for Minic, especially the 2.48 release.
So things like "the Minic net", or "Minic2.48NNUE" are not a all self descriptive. Even "Nascent Nutrient" is not "Minic net", it is just a net based on Minic data and learner. I will release more of them, with other names or with version number. There is a real need to specify what net is in use.
Some nets names are with dates, some are with SHA, some have "real" names. But in all case it should be well specified in order to maintain tracability of your great testing effort.
