to add
Game number: Main Page
White: TwixtBot    Black: David J Bush (LG) | This game (LG) | Download JTwixt file
On 2018-11-28 at 23:33, twixter (info) said:
|18.g19 looks better.

On 2018-11-29 at 11:46, BonyJordan (info) said:
|18.g19 19.f20 20.f21 21.e22 22.f16 23.i17 24.h17 25.k18 26.l15 27.h15 28.k22 29.n21 sayeth the bot

On 2018-11-29 at 16:30, twixter (info) said:
|18.g19 19.f20 20.f21 21.e22 22.f17 was the point behind G19. Prithee what sayeth the bot to that?

On 2018-11-29 at 16:59, twixter (info) said:
For example |18.g19 19.f20 20.f21 21.e22 22.f17 23.i17 24.k21 25.g18 26.e23 27.d24 28.g15 29.k18 30.m20

On 2018-11-29 at 23:10, BonyJordan (info) said:
|18.g19 19.f20 20.f21 21.e22 22.f17 23.m17 24.l17 25.k16 26.n18 27.o18 28.s10 29.o7 30.p19 and now the bot thinks Black is way ahead (+.57)

On 2018-11-30 at 17:53, purgency (info) said:
9.v4 seems like a really weird move, whats the expert opinion on this? Is this kind of move feasible?

On 2018-12-01 at 14:14, MisterCat (info) said:
If 9.v4 effectively blocks black from reaching the right, then it shows that David should NOT have swapped at move 2; instead, he could have done the same thing to TwixtBot with 2. d22 . On the other hand, if black CAN break though to the right, that would brand 9.v4 as a failure.

On 2018-12-01 at 16:53, BonyJordan (info) said:
I also thought 9.v4 was a funny looking move. Generally I'm a bit skeptical of the bot's openings, but it plays a lot better than I do at this point so who am I to say? It does look to me like White is winning in the upper right corner after 11.s7 (e.g. 12.r8 13.r9 14.t8 15.q7 and if it wasn't for the peg at v4 black wouldn't have to 15.q7).

On 2018-12-01 at 17:43, purgency (info) said:
@MisterCat Nah, if you swap then the next move isn't yours but you have the only peg in the corner. If you play 2.d22 then the next move isn't yours but you don't even have the only peg in the corner either. So the swap is for sure better than 2.d22.

9.v4 makes sense but it's really unusual to place a peg that far into a corner this early on

On 2018-12-01 at 18:41, twixter (info) said:
MisterCat, you may be right, but for now I still believe I was correct to swap. After all, the bot appears to agree with me that I would have won with |18.G19.

BonyJordan, thanks for the analysis. Your bot is great for keeping me on my toes with regard to tactics. But it seems to have underestimated my threat along the bottom. How many more "epochs" of policy refinement do you think it would take before it can win against any human right from the opening?

On 2018-12-02 at 00:40, BonyJordan (info) said:
Short answer is I don't know. Of late I am gaining call it 25 Elo points a day in self-play, but everyone knows that self-play Elo points are inflated compared to real Elo points. If you think the net plays 2000-2200 level on LG (I have no idea - it has very few losses as yet) and - making this completely up - self-play Elos are half the value of regular Elos, that would imply something like 3 months away from Elo=3000... but another thing is that nets have a maximal strength they can get to, and mine might stop before it gets that high. My current net is rather small by AlphaGo standards -- only 40 "channels" wide; AG used 192 and 256 wide and Leela is 128 wide I think.

You must be logged in to add comments.