Simulated real time learning using RomiChess

[Mike Sherwin]

Using a fresh learn.dat file two copies of RC play 100 games to 8 ply deep. Then RC uses that learn file with book_off to search for 30 seconds. Then SF13 searches the resulting position for 60 seconds using one thread. The process is then repeated for the next set of moves. I have no idea how this is going to turn out.

RC move 30 sec, SF move 60 sec, RC move without learning 60 sec.
1. e2e4 e7e5 ; e2e4

more to follow ...

[Mike Sherwin]

RC move 30 sec, SF move 60 sec, RC move without learning 60 sec.
1. e2e4 e7e5 ; e2e4
2. g1f3 b8c6 ; g1f3

No variance so far but then only best moves so far.

...

[Mike Sherwin]

RC move 30 sec, SF move 60 sec, RC move without learning 60 sec.
1. e2e4 e7e5 ; e2e4
2. g1f3 b8c6 ; g1f3
3. b1c3 g8f6 ; b1c3

b1c3 while not considered to be the most dynamic continuation is nonetheless a good move

[Mike Sherwin]

RC move 30 sec, SF move 60 sec, RC move without learning 60 sec.
1. e2e4 e7e5 ; e2e4
2. g1f3 b8c6 ; g1f3
3. b1c3 g8f6 ; b1c3
4. f1b5 f8c5 ; f1b5

Still no divergence but also no need for it, yet.

[Mike Sherwin]

RC move 30 sec, SF move 60 sec, RC move without learning 60 sec.
1. e2e4 e7e5 ; e2e4
2. g1f3 b8c6 ; g1f3
3. b1c3 g8f6 ; b1c3
4. f1b5 f8c5 ; f1b5
5. b5c6 d7c6 ; b5c6

Actually SF prefers Romi's move at depth 34!

34/49 00:35 61,284k 1,731k +0.12 Bb5xc6 d7xc6 Nf3xe5 Bc5xf2+ Ke1xf2 Qd8-d4+ Kf2-f1 Qd4xe5 d2-d3 Bc8-e6 Qd1-f3 O-O-O a2-a3 g7-g5 Qf3-g3 Qe5xg3 h2xg3 h7-h6 b2-b4 Nf6-g4 Kf1-e2 Rh8-e8 Ke2-d2 f7-f5 Bc1-b2 Ng4-f2 Rh1-e1 Nf2xe4+ Nc3xe4 f5xe4 Re1xe4 Be6-d5 Re4xe8 Rd8xe8 Ra1-f1 Re8-e6 Bb2-g7 Re6-g6 Rf1-f8+ Kc8-d7 Bg7-d4 Bd5xg2 Rf8-f7+ Kd7-d8 Bd4xa7