|
ANALYSIS ARCHIVE
Choice commentary from the list participants TURN 8:  Bob Lipton: The two moves that look attractive are d4d ENTOPIC for 22 points and 6da UNCO for 20. A little bird keeps whispering in my ear to play ENTOPIC, but I'm leaning toward UNCO. LET has to be at least two points better than LU. Doesn't it? No vote yet. Nick Deller: I'm delighted by how difficult it's getting to play on this board. Always a promising sign when you're ahead. Off the top of my head, UNCO at D6a for 20 looks like it might be the best score (I've undoubtedly missed something though), and lets in XU or XI for 31 whereas DOUCET J13a for 18 kills the bottom right for bonuses, but might allow something like OXTER as a painful reply. I think DOUCET avoids more threats than it creates though. Keeps of LET or LN aren't too disturbing at this stage. The fact that I'm happier with DOUCET might have more to do with how I play the game than which move is better. I'll keep my powder dry for a bit, largely because I want to see what I haven't spotted. Phil Nelkon: Somewhat torn here. UNCO looks good, the LET leave is particularly nice. Don't like ENTOPIC because the leave is poor by comparison. However defence is becoming a consideration. Does the play of STUD indicate opp. has last S ? If opp. plays bonus on row 14, it will score at least mid 70's which given that we will get a triple to pick off will make scores about level unless we get lucky with the Q or X. Suggest DULCET j13a for consideration, very similar to Nick's suggestion but with a better leave, that takes out most of the bonus places and I like the 5 tile turnover. I suspect the nervous ninnies will witter on about big Q & X plays, they may be right. DEUTON in the same place may be marginally better from that point of view only. Trouble with playing defensively is that we are only 40 odd ahead after such a move. Nick Deller: DULCET is almost certainly better than DOUCET - the only minor disadvantage is that DULCET would allow an -ST ending to go down where DOUCET would not. Clearly a weekend in Durham has knocked me out of my stride! I still want to keep my powder dry, but I marginally prefer DULCET to DOUCET and both ahead of UNCO. I think. I'm not sure that STUD implied that the machine has the last S - using it was necessary to contribute 15 or so points to the move score; would keeping an S ever contribute that many points to a rack leave adjustment? Despite that, I think it likely that something nasty is imminent. Stewart Holden: Agreed, its use of the S was probably for score more than any indication that it has the last one. Fortunate that both Y's have already been used, since it prevents JOE-Y, STUD-Y and BASILAR-Y without use of the last blank. Also worth noting that STUD didn't involve playing an E, and we didn't pick one up - this means that the bag is still very E-heavy with 9 out of 39 unseen tiles being an E. I think we need to block the STUD end hooks, since a bonus along that row is the machine's best chance of getting back into the game. Unfortunately, ways to block this row 14 are limited to playing a six-letter word off the D, as Nick, Phil Nelkon and others have already pointed out. The other advantage of a play such as this is that it also blocks bonus plays around the E of TREZ which people were worried about last time If someone were to work out the chances of us picking at least one E by playing five tiles at this point, I would imagine they would be extremely high - therefore DOUCET (leaving LN) would seem to be a good choice. However, opening the right-hand column is slightly worrying (fear of anything like EXTRA for 60) and so I was wondering about just DOUC instead. DOUC leaves us with ELNT (reasonable), blocks row 14 and still goes a long way towards blocking the E of TREZ. Then again, it only scores 14 and the risk of an EXTRA-type move isn't too great. I'll put my 15 votes on DOUCET here and now to get the ball rolling. Keiichiro Hirai: Now I vote (17votes) for COZEN, keeping LUT. Machine's play of "STUD" seems to indicate that it has another 'S' or a blank. So, in our turn, decreasing Maven's bingo possiblities has the big priority, I think. Bingos on Columns N and O are most probable like -IZE, and using E of TREZ. COZEN can kill ENQUIRE(D,R,S) possibility, too. (But CLOZE cannot.) Getting rid of C&N and keeping the last L is good. Bingos played under STUD on Row14 by Maven is not a big threat. That would give us chance to score on TLS. Other considerations: DEUTON(14) at J13a, keeping CL DOUCET, DULCET may be countered by QUEER or QUEEN, etc. CENTO/OHM(17) keeping LU UNCO/OHM(20) keeping LET ZONULE looks dangerous Stewart Holden: Can't agree with that logic at all, sorry. Row 14 is by far the biggest threat to us at the moment, compared to the relatively small chance of a bonus around the Z. With regard to the last move, I am also surprised by the sim results. I totally agree with Pete (and others who've said the same) that FOE would appear to be much better than the moves which placed above it. I'm going to run a very comprehensive simulation on that move when I'm back on my fast computer in Exeter, but that won't be for about three weeks. I'm sure if we look far enough ahead, FOE will come out better. Simulations are limited in accuracy, of course, by the number of moves ahead they consider and the number of iterations. The machine has a little-used feature called "End-analyser" whereby, if the number of look-ahead moves is set to something like 100, it will play right through to the end of the game for each iteration. This, I believe, makes tile turnover and other factors more relevant to the final results. The famous situation from the 1995 WSC final is a good example of simulation difficulty. I can't find the board position easily, but it's the LADYBUGS / BEADY / PEJORATE situation which I'm sure many people will have seen before. David Boys could have played LADYBUGS for 89, but saw the tiny risk of Joel Sherman playing PEJORATE for 212 and so blocked it with BEADY for 45 instead. A simulation of that position using the machine will _always_ tell you that LADYBUGS is the best move, since it was the only playable bonus and therefore gained, on average, more points over the next few moves than anything else. However, it is also undeniable that BEADY was a better move; the playable bonus gave him a larger average winning spread, but would only have won, say, 99.5% of the time in that situation. By blocking PEJORATE he was guaranteed to win every time, but by a smaller spread. In the final game of the WSC final, of course, spread doesn't matter in the slightest and so the 100% winning move must be better. What I'm saying is that the machine's simulations do not take into account the percentage of games which will be won by playing a certain move. Using this fact in Move 8, I can argue that DOUCET/DULCET/DEUTON or a similar move is probably a much better play than any simulation will tell you. Plays such as ENTOPIC or UNCO will score more points on average because the machine is not always going to play a bonus under STUD, through the E or Z of TREZ or anywhere else. However, on the occasions when it does, we're going have a greatly reduced chance of winning the game. A six-letter play from the D of STUD can be likened, therefore, to the BEADY play. A simulation won't agree, I know it won't, but something like DOUCET will win the game for us a greater percentage of the time than anything else. Nick Deller: This argument is the same one I would make about FUZEE. I still think it might come out on top in terms of win %age. Giving away 7 points of raw score against FOE, it only simulated about half a point worse, so it must be significantly better defensively - and that can only be down to the EZ floaters that it's successfully defending against. I'm encouraged rather than downheartened by the basic simulation results. I have a question about these simulations, by the way; is the machine being given a completely random initial rack each time, or is it being given its unknown - but fixed - retention from last time plus a fresh draw of tiles? It would make a difference when our choice of move is partially based on a correct deduction about what the computer kept the previous turn. Darryl Francis: It seems to me that we are vulnerable to bonuses from the opponent in five places: 1. the O hook for (O)HM; 2. the S hook for STUD(S); 3. through the floating T of TREZ; 4. through the floating E of TREZ; 5. through the floating Z of TREZ. We can't block all 5, but we can block 4 of these, in the bottom righthand corner. Someone has already suggested DOUCET and DULCET. DOCENT and DOLENT are also possibilities. All of these suffer from a strong Q play at N10. Of these, I prefer DULCET, retaining the O for the (O)HM possibility. Unfortunately, this uses up a U, and the Q is still to come. Alternatively, we could play a 5-letter word from the D of STUD, such as DOUCE, DOLCE, DONUT, etc. But all suffer from taking extensions at O13, giving the opponent access to the TWS at O15. However, many of these would force him to burn an S, if he has one. My favoured move at the moment is to play DULCET. I don't like the suggested UNCO play at 6Da, because of the possibility of an X-play by the opponent on row 5. No vote yet. Just thinking out loud. Pete Finley: UNCO scores more than any of the defensive D+ words we're considering, but a defensive move has to be right now in terms of improving our chances of winning the game. I don't care what the sim says. But which move? Is the threat of the Q on N10 a bigger problem than giving away the tws? I believe DEUTON does most to minimise the N10 threat, but the X under the O could be costly - although not as bad as the Q at N10. I don't much like the CL leave either. However, all the other options put either an E or an N in the line of fire of the Q, or take an E or S extension in that line. Not confident about it, but I'm going with DEUTON. It won't sim well, I know that, because it doesn't score enough compared to the others, but it's the move I'm happiest with in terms of winning the game. Ray Tate: Way behind schedule!! Can't see any viable alternative to UNCO D6a (20) with nice ELT leave, which even with a 2 E pick-up from those missing (9 E's contained in 39 unseen tiles) seems OK. Can't see any effective ways to take out the L, N & O files or 14th row with anything meaningful but again we should get counterplay if the machine bonuses. Not worried about SQUIB from B2d or parallel X-plays like MAXI, TAXA, TAXI from B5. 16 votes UNCO D6a in meantime. David Webb: The leave-adjusted values of some candidate moves are:
|
Copyright © 1999-2000 by Mind Sports Organisation Worldwide Ltd.
E-mail: info@msoworld.com