A missed super-criss-cross squeeze and Some random notes

Interesting bridge hand from today.

Dummy: S:T9xx H:Txx D:KJx C:AJx

Declarer S:AKQ H:AQ9xx D:Ax C:Txx

Contract 4 Hearts with the lead of the spade 7 from West.

Short story — winning, I decided not to take the diamond hook, since it was 50/50 and there may be a stiff spade. So Diamond ace, diamond king, small heart (catering to KJ tight) When the jack appeared without much thought and the queen won, I decided to drop the king and use the HT as an entry, which worked. At this point pleased with the results so far I forget to rectify the count  for a squeeze by ducking a club. I didn’t, and only made five (when spades were 4-2 and nothing particularly good happened).  I made a comment “there must be some Criss-Cross squeeze somewhere” (see link for details). So I was awake and yet brain dead at the same time. Typical.

Later on I realized I messed up even earlier. Given that I am not taking the diamond finesse, better to keep the ace in my hand and keep Criss-Cross positions in all three suits! This also has the benefit when I lose a trump trick that I may not get overruffed in diamonds (rare) or lose my DJ menace (common). Now when the hearts break I can lead a club right away (I need to lead the Jack) and then win the return and draw trumps. Assuming a spade return, I end up with

Dummy S:T9 D:Jx C:A

Hand S:Q D:A C:Tx

There has to be a squeeze here (unless a card is already falling under the ace). Since each suit is blocked, its a kind of super-criss-cross. Assume one player guards spades (Jx) and Diamonds (Qx). I lead the club ace and then it resolves to a criss-cross with spades and diamonds. If I think that one player guards spades and clubs, I lead the diamond ace, and if I think one player guards clubs and diamonds I lead the spade queen. (Of course, if one helpless soul guards all three, they have already been squeezed).

As in any criss cross, declarer must read the position correctly: not likely given that I botched the hand. (And a non-spade return also breaks up the criss-cross in that suit, so I may have to go to a less esoteric line). Still, I don’t think I’ve seen this before, but probably I just never noticed before.

(Update — As Jeff pointed out in the comments, this is all wrong).

I am also resolved to not bid any more grand slams in club games, even if I think it is highly likely. Only one pair bid the correct small slam, so even if the grand made (sigh) I was risking 9 matchpoints to gain 1. The hand was a rock crusher S:AKQJ H:Axx D:AKxx C:Ax.

Also played today — More Jump Drive, and Vast: the Crystal Caverns. Regarding the latter — some good ideas, nice asymmetry between the players, although it sounds like they aren’t well balanced. Good art, somewhere between Groo and The Grizzled. The TaoLing also played Illumat, which I must admit looks wonderfully gothic.

Round numbers

Just played my 111th game of Sentinels and logged my 777th session of bridge yesterday (undoubtedly low, but round numbers are round).

A bit of Wild Blue Yonder campaign & thoughts

As advertised, I set up the easiest campaign (Rommel Attacks) and played both sides through the first mission. If you’ve played Down in Flames (or just skimmed my review) you’ll have a rough idea of how dogfighting works. The simplest campaigns are Land campaigns, which comprises 3+ missions. Here’s how it works (after you pick the campaign and sides), by walking through the first mission:

1. You draw a card to get your target and starting aircraft. Since I was playing Stage I of Rommel Attacks, that campaign had a chart and said the Axis were trying to bomb an airfield at Gazalla. The airfield is the target, and it shows flak ratings (both area flak, and over target), and how much damage each hit, direct hit or vital hit (from the bombs) does to the airfield, and the VPs for various damage levels to the airfield. The scenario noted that Gazalla had less flak than a typical airfield.
2. The card drawn also shows the base airplanes for each side. In this case, the Luftwaffe had a leader + a wingman in a Bf 110 E, and the allies had … nothing.
3. Now, each player gets to pick a resource. There are 3 resources (and three mission) so you get to pick each one once. Each of the Allied resources includes a Hurricane Mk I Leader + Wingman. The allies took that (and nothing else) and the axis took a Bf 110C.
4. The scenario sets the duration — 9 “target-bound” turns, and 7 “home-bound” turns. (There’s always an ‘over-target’ turn in between). (In reading the rules again, I missed something. This is the base time, but is reduced by the speed of the slowest attacker … so instead of 9-1-7, it should be 3-1-1. A big change!)
5. If your planes have special pilots (as a resource) you get them, otherwise you draw for bonus pilots. The axis 110E’s wingman is good, so he gets +1 Performance (basically an extra offensive card in his mini-hand). To balance, good pilots that get shot down cost extra VPs.
6. The 110E’s are loaded with bombs (the extra planes are escorts). There are a number of penalties for “loaded” planes, but the most important is this: You can’t inspect your hand until you need it, and you can’t attack unloaded planes unless they began their turn engaged with them. So the Bf 110 E’s can’t target the Hurricanes until they drop their bombs (or jettison them, which costs VP). Loaded planes also can’t play some cards (Full Throttle, Half Loop or Scissors) and wingmen have reduced offensive rating.
7. The airplanes all choose their altitudes (depends on the mission), but I just picked medium for everyone.

So, here we go. The German’s Bf 110E’s (simply “E”) are loaded, so that hand goes face down. The Axis go first and the Bf 110 C (“C”) wingman doesn’t go (the very first wingman skips their turn). The leader tries a simple burst and fails, having a hand full of only attacks. He discards another card to replace.

The Hurricane’s Wingman (“H”-W) attacks the C’s leader. The hurricanes could maneuver to attack the loaded E’s, at which point the C’s could attempt to intervene to cancel the attack, or the E’s could take it and pick up their hand. This would let the E’s also start discarding and redrawing, but that’s skipped while the hand is face down, so in my judgement the Hurricane’s should focus on the escort for a while. The wingman gets a small burst and hits. The H-leader fires a more dangers burst at C-L, and now C-L responds an cancels. The E’s turn is skipped. The next Target Bound Turn (you count down) has some maneuvering (all cancelled) and the hurricane’s do a bit more damage. Both sides have their maneuvers cancelled, but eventually the Hurricane’s get the advantage and damage the C-Leader. Another two turns of dogfighting and the C-leader is down in flames, promoting the C-wingman.

All this works exactly as the dogfight game. If the Hurricane’s target the E’s, then intervention applies.

The rest of the Target bound turns see the Hurricane’s trying to finish the escort. (Part of that was due to my misreading of the VP rules). Finally, at the end of Target Bound -1, some special rules apply. Normally, only aircraft planning on bombing go over target, since there is flak. The E’s get an “over target” marker and the other planes don’t.  The Bf 110E’s take some minimal area flak (only card against each, but you can’t respond!) but are in luck — no damage.

During the over-target turn, the planes not over the target dogfight normally. The scrappy C-leader (former wingman) gets a cockpit hit on the Hurricane wingman, which doesn’t do much damage, but keeps him from drawing any more offensive cards. Nice! But the Hurricane leader damages the C wingman, so he flips over. The E’s take their turn, they drop to low (required to bomb) and then draw a mini bombing hand, based on their altitude, target and plane bomb rating.

Each card has a four bomb results “Miss,” “Hit,” “Direct,” and “Vital.” In order each plane draws its hand, and then British draws a flak hand. Flak can do two things: Some cards “Spoil an attack” which turns the best opponents card into a miss. The cards may also do damage, but can be responded to. But … the attacker is considered loaded when responding, and will also reduce one of their hits one level).

So — a minimal “mini hand” is drawn for each side and played out, mostly luck, but possibly some decisions (respond and lower your damage, or not). The mini hands I draw have no options, since the minimal flak does not spoil any attacks or even damage the planes. The Germans do enough damage to earn 3 VPs, but are down five for losing a plane and two more for a damaged plane. The E’s then have to take area flak (which again, is just a draw they can’t do anything about) on their way out and get a scrape or two but nothing serious.

The next turn the hurricanes try to finish the remaining Bf 110C but narrowly fail. The E’s are now unloaded and they climb up and engage the hurricanes. The hurricanes take one last shot at the damaged C and then shrug and climb to high, then vertical roll to very high, above the German planes’ ceiling. The Hurricanes could try to voluntarily disengage (there’s no VP penalty for this during non-dogfight missions) but that’s risky and by going very high they can effectively time out the game, so that ends it.

VP are mission VP (3 VP for the damaged airfield for the Axis) and then damaged planes (7 VP for the allies). But there’s a somewhat confusing rule I didn’t notice earlier that could be interpreted a few ways, but in this particular case every reading gives the Axis 4 VPs since both planes struck the target and ended undamaged, so that’s 4 more VP for the Axis.

It’s a draw!

Of course, If I had realized that I got the number of turns wrong and the VP bonus for each plane that bombs the target, the Hurricanes may have decided to charge the bombers.. Ah well, I played it solo to get the rules right eventually, not right “This time.”

Now — the campaign missions could get more complex than this. Rules I haven’t touched include:

• Formation bombers — these act like wingman and typically can only defend (and bomb). This leads to a lot more
• Intervention — It could have happened this game, but it didn’t. Perhaps that was wrong on the part of the Hurricanes.
• Fuel — This is an optional rule.
• Special rules — some campaigns have a page or so of rules, or some missions that may show up that have special rules.

There are also progressive campaigns, where you keep the same pool of aircraft between missions, deal with weather, and don’t have a fixed number of missions, but VPs until either it times out or one side wins.

Anyway, it will take a few more games to work the kinks out but the game didn’t take too long (even though I played 11-13 turns instead of 5!, and recorded some notes), so I think this is playable once you are willing to invest the time….

Some thoughts

1. I picked up Civ VI as my holiday computer game (counting Portal Bridge Constructor as my pre-holiday computer game). $30 until end of the year. I like, but I prefer the old, easy to read graphics. Also, the tutorial can be multi-hour and has no save. Sheesh. But the split of research into science (technology), civics (governments and improvements) and faith (religion) plus great leaders all having their own points seems correct. It may be overgilding the lily.
2. Speaking of, I’m about 40% of the way through P.B.C., fine for a game that cost $10. It’s difficult already.
3. The Wild Blue Yonder Solitaire campaign does not capture any of the joy of the game. I’ll probably play both sides in an actual campaign to learn the rules (I hope). Not great solitaire, since its an action-response basis and I’ll know both hands (somewhat); acceptable as a rules exercise.
4. I may write a Year In Review, but this wasn’t a big year for gaming (for me).

Happy Holidays!

Follow up, links

Re: Alpha Zero — Thoughts from Ken Regan, including how this affects his thoughts that a perfect chess player would be rated ~3600 ELO, and how to generalize the algorithm to hard (non-game) problems by making the problem an opponent.

Re: Universal Paperclips — Ted Chiang (author of the story that became Arrival) on how the “Paperclip AI” that destroys the world by trying to maximize a single parameter already exists …. and is called ‘the corporation.’

As a Factorio player I’m not in much of a position to laugh at Minecraft …. (and the SouthPark episode “Informative Murder Porn” had some great Minecraft gags), but at least I can say that my game didn’t inspire a truly massive botnet that disrupted much of the US Internet last year.

And I’m taking a small bite-of-cake-sized break from Factorio to spend some time on a new game with an old friend. For Science. You Monster.

Wild Blue Yonder Initial Thoughts

I finally got to play a pair of 2 v 2 dogfights for Wild Blue Yonder, so a few thoughts. I poked around for a review for Rise of the Luftwaffe I wrote years back and found it on BGG, and not much has changed, for the dogfight. Simply, during your turn you pick an enemy (if you aren’t already in a furball) and play an attack card (which may give you position, or actually fire your weapon, or do a few other things). Your target may respond. You may respond to his response, etc. If the ‘attacker’ played the last response, the attack card takes effect. If the defender did it’s cancelled, but you can play as many attack cards as you qualify for (actually attacks cost “bursts” which are limited by your plane and position, but maneuvers are typically always playable, unless you already tailing your target).

Since most response cards cancel 2-3 cards, there’s probably a decent amount of skill of knowing the best way to cancel. You can also count cards. On the bigger picture you have to decide when to target a leader or the wingman, and play with altitude. I suppose its not surprising there are no strategy articles on BGG. It seems so simple, but I suspect even the dogfight has real depth. Yes, its a card game, so even a bad player can win sometimes, but skill will usually show in an equal match. (I do think the supposedly equal match (based on card VPs) that I offered was in fact slightly unbalanced, the British planes extra HP being less valuable than the better hand size for the Germans. The Germans won handily against the novice British, but the second game was much closer, although still a German victory, but with significant losses).

Unlike prior entries in the series, each side has their own deck of cards, which means if you draw the very good Ace Pilot cards (cancel any other card), I may still have mine in my deck. Ditto any other good or bad card. They may still cycle out in a bad time (when a wing-man draws their mini-hand) but at least its not “He got aces and I got deuces.” Your aces may show up later — possibly too late, but statistically it is more likely to even out. A nice touch.

I’ve read the campaign rules and would like to play a campaign. There was a solo campaign promised but it doesn’t seem to be anything like the regular ones. I guess I should have expected that, given the complexity of the card game, but I’m disappointed. There are many, many campaigns. The rules don’t seem complex, but they are numerous.

I was ready to really praise the insert. As some of you may know, I typically just chuck inserts. This seemed great. It could hold the cards while storing most counters underneath (although bagging the campaign pilots to keep them out of the way). But when I put the actual airplane cards in I discovered the space was so tight I couldn’t easily get them out, even with a thin blade I could only pry up 1/2 at a time. (When they were shrink wrapped pulling one out pulled them all out). Insert tossed. The box is the nice and sturdy type GMT has. I haven’t sleeved my cards, but I may if I play it frequently. (The cards seem generally nicer than the ones I remember, but its been a few years since I traded my copy away).

I think even interested ten year olds could play the dogfight. I’m hoping to interest the TaoLing in the campaign at some point. We’ll see.

Pay no attention…

Factorio 0.16, (the final pre-release, improved graphics and optimizations and a few new features) dropped this week. Expect nothing and ye shall receive.

But I got my copy of Wild Blue Yonder in the mail, so I may actually play that at some point as well, so expect something and ye may receive.

 

I’d never considered ordering from Cool Mini Or Not ….

Since I don’t really buy minis. Imagine my surprise that I’ve placed an order with them. Especially since (AFAIK) I’ve never told them my (admittedly not hard to find email address) or even visited their website.

OK, I didn’t place an order. Someone else did. The invoice gives me a name, shipping address, a partially redacted CC number etc.

So, I tried to do the right thing and email their support address (given in the invoice) to say “Hey, maybe you don’t want to send your customers information to random people”

And … no, apparently the email that was generated a few hours ago has out of date information.

No, I will not call you up or order a ticket to fix your problem.

So — I would strongly advise against doing any business with them. They’ve earned it.  At worst? Some spam marketing thing or phishing. Best case? Mishandling customer data

Personal memo to K__ V______. Enjoy your Shael Hon stuff. Whatever that is.

Re: Mastering Chess and Shogi by Self-Play (etc etc)

More stunning news from the DeepMind crew:

In this paper, we generalise this approach into a single AlphaZero algorithm …. Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi (Japanese chess) as well as Go, and convincingly defeated a world-champion program in each case. –Abstract

And how did they do?

In chess, AlphaZero outperformed Stockfish after just 4 hours; in shogi, AlphaZero outperformed Elmo after less than 2 hours; and in Go, AlphaZero outperformed AlphaGo Lee after 8 hours.

(Stockfish and Elmo are the best computer programs in the world. AlphaGo Lee beat Lee Sedol, but was later improved on by AlphaGo Zero).

And after 3 days, AlphaZero played a 100 game match against Stockfish and won 28, drew 72 and lost zero.

It did this in spite of a slower search speed:

AlphaZero searches just 80 thousand positions per second in chess and 40 thousand in shogi, compared to 70 million for Stockfish and 35 million for Elmo. AlphaZero compensates for the lower number of evaluations by using its deep neural network to focus much more selectively on the most promising variations – arguably a more “human-like” approach to search

The full paper is worth a read. I don’t understand all the details, although its non-linear evaluation function apparently may have biases (like a hand crafted evaluation function) but the authors argue that their Monte-Carlo search (with its inherent randomness) will “average over [AlphaZero’s] approximation errors” but the traditional Alpha-Beta search propagates them.

Looking at some of the commentary by more experienced chess players, this jumped out at me:

The paper also came accompanied by ten games to share the results. It needs to be said that these are very different from the usual fare of engine games. If Karpov had been a chess engine, he might have been called AlphaZero. There is a relentless positional boa constrictor approach that is simply unheard of. Modern chess engines are focused on activity, and have special safeguards to avoid blocked positions as they have no understanding of them and often find themselves in a dead end before they realize it. AlphaZero has no such prejudices or issues, and seems to thrive on snuffing out the opponent’s play. It is singularly impressive, and what is astonishing is how it is able to also find tactics that the engines seem blind to….

In this position from Game 5 of the ten published, this position arose after move 20…Kh8. The completely disjointed array of Black’s pieces is striking, and AlphaZero came up with the fantastic 21.Bg5!! After analyzing it and the consequences, there is no question this is the killer move here, and while my laptop cannot produce 70 million positions per second, I gave it to Houdini 6.02 with 9 million positions per second. It analyzed it for one full hour and was unable to find 21.Bg5!!

(Emphasis mine, also to note that AlphaZero and stockfish had 1 minute per move).

The DeepMind team just keeps delivering stunning advances. Someone is probably jumping with a reasonable approximation of joy.

The Latest

I’d been waiting for Magic Maze to show up at the game store. It never did. However — Barnes and Noble got it, and the TaoLing’s class was having a fundraiser there (anything you but they get 10%), so I got it. B&N has a shockingly good games section. Way too many tie-ins … (seriously, Game of Thrones Settlers of Catan? The pure-non-conflict Euro meets the Red Wedding? Sheesh)  but there are jewels there.

Anyway, after a few games 2 player, I came up with an obvious variant:

Use the four player tiles and rotate them normally, so you only change 1/2 of what you do each rotation (instead of having the same things always grouped). For more weirdness, assign a tile to each hand, and then you have to do the actions with the correct hand.

Magic Maze isn’t great two player, but its not necessarily easier. We managed to lose several times, but now we’re at the “full game” (Scenario 7).

Haven’t tried solo Magic Maze. Doesn’t seem like it would be goodl

After a few weeks off, picked up my Factorio Mega-base again. Coming along nicely. Last time I posted I was at 20 rockets at 55 hours. I improved smelting, got more oil refining. This required breaking the 1 Gigawatt barrier and an hour or two of tactical nuclear devices to pacify the natives and clear out more area for mining.  Anyway, I’m at around 85 hours and 210 rockets. Always something to improve.