JJJ wrote:And here is another elo jump in selfplay with the ID 237. 150 elo overall in less than a week folks.
Nah, this is self-play with fixed number of playouts. But the nets became slower. At fixed time against standard engine, there was little progress lately, even with the introduction of the larger net, and only on GPU, not on CPU (but CPU results will be anyway irrelevant). On GPU, with the introduction of ID227 larger 15x192 net, maybe there was some improvement, 20-30 Elo points or so, no more, because the nets were much slower. But since larger nets, from ID227 to ID237, in their 500,000 games, I cannot measure sensible improvement at fixed time against a standard engine, probably less than 15 Elo points. This is somehow strange, as the nets are fresh to learn.
For curiosity, I plotted a graph of LC0 at 1 playout pitted against Komodo 11.3.1 at depth=1. You have to keep in mind that this is not fixed time, this is more comparing the evaluations, and I took Komodo because it is among the strongest standard engines at depth=1.
The large, sudden increases are the introduction of larger nets: at ID123 and ID227. Each datapoint is 1000 games, and ID237 shows some small progress compared to ID227, but it is offset somehow by maybe lower speed by 3% or so in fixed time games.
.
