Life In 19x19 http://lifein19x19.com/ |
|
Thermography http://lifein19x19.com/viewtopic.php?f=12&t=17788 |
Page 9 of 19 |
Author: | Gérard TAILLE [ Thu Oct 29, 2020 1:18 pm ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Gérard TAILLE wrote: When you use a positive tax, say tax = 9, you keep the value 1 of the prisoner or you change to the value 9 ? Yes, because with a positive tax you want to reduce the value of a move for the purpose of finding the mean value of a game. I am a little lost Bill. When you answered "yes" it concerns "you keep the value 1 of the prisoner" or "you change to the value 9" ? |
Author: | Bill Spight [ Thu Oct 29, 2020 1:37 pm ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Bill Spight wrote: Gérard TAILLE wrote: When you use a positive tax, say tax = 9, you keep the value 1 of the prisoner or you change to the value 9 ? Yes, because with a positive tax you want to reduce the value of a move for the purpose of finding the mean value of a game. I am a little lost Bill. When you answered "yes" it concerns "you keep the value 1 of the prisoner" or "you change to the value 9" ? Sorry. It has been, and still is, the practice in thermography to keep actual scores the same before applying the tax. I only suggested changing the value of a prisoner to illustrate a point about a tax < -1. I was not endorsing using such a tax. |
Author: | Gérard TAILLE [ Thu Oct 29, 2020 2:36 pm ] |
Post subject: | Re: Thermography |
BTW Bill just a question about notation. When you use slash notation like {5|2} how do you indicate it concerns a ko which can be connected say by black? |
Author: | Bill Spight [ Thu Oct 29, 2020 3:20 pm ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: BTW Bill just a question about notation. When you use slash notation like {5|2} how do you indicate it concerns a ko which can be connected say by black? Here is a good way to indicate a ko. K = {1||K|0} |
Author: | Gérard TAILLE [ Thu Oct 29, 2020 3:33 pm ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Gérard TAILLE wrote: BTW Bill just a question about notation. When you use slash notation like {5|2} how do you indicate it concerns a ko which can be connected say by black? Here is a good way to indicate a ko. K = {1||K|0} Thank you Bill |
Author: | Gérard TAILLE [ Sat Oct 31, 2020 3:35 am ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Here is an sgf file for Three-Points-Without-Capturing, assuming no ko threats. ![]() I am completly lost Bill. Let's come back to the beginning Bill Spight wrote: Here is the game tree: {-3|-3||-2}. How do you find this game tree? Why not {-3|-2} or {-3|-3||-2|-2} or some longueur notation to take into account black and white sequences? Bill Spight wrote: This is a standoff, as neither player wishes to play first OK Bill, no problem Bill Spight wrote: In CGT this reduces to { |-2} Why not {-3|-2} Bill Spight wrote: To handle such games Berlekamp extended thermography downward to temperature -1, which he called subterranean thermography. Below temperature 0 the vertical blue scaffold shows up. I do not see how such games can be handled without more information than the game tree. To take another example, what will be the game score of the game tree {-20|-2}? You cannot caculate it can you? |
Author: | Bill Spight [ Sat Oct 31, 2020 10:26 am ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Bill Spight wrote: Here is an sgf file for Three-Points-Without-Capturing, assuming no ko threats. ![]() I am completly lost Bill. Let's come back to the beginning Bill Spight wrote: Here is the game tree: {-3|-3||-2}. How do you find this game tree? Why not {-3|-2} or {-3|-3||-2|-2} or some longueur notation to take into account black and white sequences? Sorry. I have revised the game tree to show why the result is {-3|-3} when Black plays first. The reason is that there is a dame left which either player can fill. (When White plays first there is no dame which either player can fill.) Normally, go players ignore the dame for territory scoring, because they do not alter the final score. For thermography they do not alter the result at or above temperature 0, so it is convenient to ignore them as well. However, they do matter when the original position is itself a number (score). Which means that they do matter for subterranean thermography. ![]() We could show a large game tree with branches at each turn, but the long sequences down to the dame form units. We could show that fact by backing up the game tree, but that would be tedious. Quote: Bill Spight wrote: This is a standoff, as neither player wishes to play first OK Bill, no problem. Seeing that neither player wishes to play first is the key to seeing that the long sequences are units. ![]() Quote: Bill Spight wrote: In CGT this reduces to { |-2} Why not {-3|-2} Because {-3|-3} does not reduce to -3. Sure, go players normally ignore dame for territory scoring, but they do matter for CGT. In terms of CGT the prototypical value of -3 is when White has 3 Black prisoners, with prisoner return. I'll come back to prisoner return shortly. In this case Black has 2 prisoners and White has 5, for a net result of 3 prisoners for White. That fact is represented by the number, -3. With prisoner return does Black have a play? Yes, Black can return a prisoner so that Black has 1 prisoner and White has 5, for a net score of -4. Similarly, White can return a prisoner for a net score of -2. I.e., -3 = {-4|-2} which is different from {-3|-3}. ![]() Now, the simplest tree for -3 is { |-2}. For instance, suppose that White has 3 prisoners, but Black has none. Then Black has no play, but White can return 1 prisoner for a score of -2. I.e., -3 = { |-2} The question is then {-3|-3||-2} =?= { |-2} = -3 ; i.e.: {-3|-3||-2} + 3 =?= 0 In CGT 0 is defined as a game in which the second player wins. As with Nim, a player wins when her opponent has no play. In its simplest form 0 = { | } I.e., neither player has a play, so the first player loses and the second player wins. Let's play the left side out. If the second player wins, it is equal to 0. ![]() 1) Black to play plays to {-3|-3} + 3 Then White replies to -3 + 3 = 0 which is a win for White, the second player. 2) Black returns a prisoner, for {-3|-3||-2} + 2 Then White replies to -2 + 2 = 0 and wins. Now let White play first. 3) White to play plays to -2 + 3 = 1 Which, OC, is a Black win. So the second player wins and {-3|-3||-2} + 3 = 0 , and {-3|-3||-2} = { |-2} Quote: Bill Spight wrote: To handle such games Berlekamp extended thermography downward to temperature -1, which he called subterranean thermography. Below temperature 0 the vertical blue scaffold shows up. I do not see how such games can be handled without more information than the game tree. To take another example, what will be the game score of the game tree {-20|-2}? You cannot caculate it can you? Sure. ![]() {-20|-2} = -3 Let's play out {-20|-2} + 3 1) Black first can play to -20 + 3 = -17 which is obviously a White win. 2) Black first can play to {-20|-2} + 2 Then White plays to -2 + 2 = 0 and wins. 3) White first can play to -2 + 3 = 1 which is a Black win. So {-20|-2} + 3 = 0 , and {-20|-2} = -3 ![]() Edit: The thing is, {-20|-2} is already has a score. Neither player has to play if the game is equal to a number, i.e., a score. Hence, Three-Points-without-Capturing. White does not have to capture to get 3 points. Interesting that go players also said that {-3|-3||-2} = -3 At least, until 1989. ![]() |
Author: | Gérard TAILLE [ Sat Oct 31, 2020 10:52 am ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Gérard TAILLE wrote: Bill Spight wrote: Here is an sgf file for Three-Points-Without-Capturing, assuming no ko threats. ![]() I am completly lost Bill. Let's come back to the beginning Bill Spight wrote: Here is the game tree: {-3|-3||-2}. How do you find this game tree? Why not {-3|-2} or {-3|-3||-2|-2} or some longueur notation to take into account black and white sequences? Sorry. I have revised the game tree to show why the result is {-3|-3} when Black plays first. The reason is that there is a dame left which either player can fill. (When White plays first there is no dame which either player can fill.) Normally, go players ignore the dame for territory scoring, because they do not alter the final score. For thermography they do not alter the result at or above temperature 0, so it is convenient to ignore them as well. However, they do matter when the original position is itself a number (score). Which means that they do matter for subterranean thermography. ![]() We could show a large game tree with branches at each turn, but the long sequences down to the dame form units. We could show that fact by backing up the game tree, but that would be tedious. Quote: Bill Spight wrote: This is a standoff, as neither player wishes to play first OK Bill, no problem. Seeing that neither player wishes to play first is the key to seeing that the long sequences are units. ![]() Quote: Bill Spight wrote: In CGT this reduces to { |-2} Why not {-3|-2} Because {-3|-3} does not reduce to -3. Sure, go players normally ignore dame for territory scoring, but they do matter for CGT. In terms of CGT the prototypical value of -3 is when White has 3 Black prisoners, with prisoner return. I'll come back to prisoner return shortly. In this case Black has 2 prisoners and White has 5, for a net result of 3 prisoners for White. That fact is represented by the number, -3. With prisoner return does Black have a play? Yes, Black can return a prisoner so that Black has 1 prisoner and White has 5, for a net score of -4. Similarly, White can return a prisoner for a net score of -2. I.e., -3 = {-4|-2} which is different from {-3|-3}. ![]() Now, the simplest tree for -3 is { |-2}. For instance, suppose that White has 3 prisoners, but Black has none. Then Black has no play, but White can return 1 prisoner for a score of -2. I.e., -3 = { |-2} The question is then {-3|-3||-2} =?= { |-2} = -3 ; i.e.: {-3|-3||-2} + 3 =?= 0 In CGT 0 is defined as a game in which the second player wins. As with Nim, a player wins when her opponent has no play. In its simplest form 0 = { | } I.e., neither player has a play, so the first player loses and the second player wins. Let's play the left side out. If the second player wins, it is equal to 0. ![]() 1) Black to play plays to {-3|-3} + 3 Then White replies to -3 + 3 = 0 which is a win for White, the second player. 2) Black returns a prisoner, for {-3|-3||-2} + 2 Then White replies to -2 + 2 = 0 and wins. Now let White play first. 3) White to play plays to -2 + 3 = 1 Which, OC, is a Black win. So the second player wins and {-3|-3||-2} + 3 = 0 , and {-3|-3||-2} = { |-2} Quote: Bill Spight wrote: To handle such games Berlekamp extended thermography downward to temperature -1, which he called subterranean thermography. Below temperature 0 the vertical blue scaffold shows up. I do not see how such games can be handled without more information than the game tree. To take another example, what will be the game score of the game tree {-20|-2}? You cannot caculate it can you? Sure. ![]() {-20|-2} = -3 Let's play out {-20|-2} + 3 1) Black first can play to -20 + 3 = -17 which is obviously a White win. 2) Black first can play to {-20|-2} + 2 Then White plays to -2 + 2 = 0 and wins. 3) White first can play to -2 + 3 = 1 which is a Black win. So {-20|-2} + 3 = 0 , and {-20|-2} = -3 ![]() Interesting that go players also said that {-3|-3||-2} = -3 At least, until 1989. ![]() Let's take an example of position with tree game {-20|-2} with an obvious seki the score of the game is -12 isn't it? |
Author: | Bill Spight [ Sat Oct 31, 2020 11:03 am ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Let's take an example of position with tree game {-20|-2} with an obvious seki the score of the game is -12 isn't it? Well, counting the marked area I get a net local score of -12. ![]() How do you get {-20|-2}? Edit: The seki has the following game tree. {8|-8||10|-6} = 0 Obviously, it is a second player win. In the marked area, White to play will choose to fill in one point of territory instead of playing in the seki, and Black will choose to play a dead stone inside White's territory. That gives us this game tree for the marked area. {-13|-11} |
Author: | Gérard TAILLE [ Sat Oct 31, 2020 11:28 am ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Gérard TAILLE wrote: Let's take an example of position with tree game {-20|-2} with an obvious seki the score of the game is -12 isn't it? Well, counting the marked area I get a net local score of -12. ![]() How do you get {-20|-2}? easy I considered only ( ![]() Now I begin to see what you mean but I have to take some time to study your analysis. Thank you Bill |
Author: | Gérard TAILLE [ Tue Nov 03, 2020 6:15 am ] |
Post subject: | Re: Thermography |
I progress in CGT but I have a basic question. We found a number of examples of uncomparable positions. My question is the following : let's take two positions P1 and P2, each one having a miai value less or equal to 1. Can P1 and P2 be uncomparable? If yes I am interested by an example If no I am interested by a clue for the proof |
Author: | Bill Spight [ Tue Nov 03, 2020 9:31 am ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: I progress in CGT but I have a basic question. We found a number of examples of uncomparable positions. My question is the following : let's take two positions P1 and P2, each one having a miai value less or equal to 1. Can P1 and P2 be uncomparable? If yes I am interested by an example If no I am interested by a clue for the proof Positions where the top plays have miai value of 1 (Edit: Corrected typo.) can indeed be incomparable. For instance, {2|0} and {3|1||0}. See https://senseis.xmp.net/?GoInfinitesimals . Also, {3|3} is incomparable with 3, strictly speaking. ![]() However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Chilling is derived there, but the derivation uses aspects of CGT with which I am not familiar. FWIW, it seems to me that chilling works mainly because scores in regular go are numbers. Also, because the values of positions with miai values less than 1 are bounded by the nearest integers. As David Wolfe said, you can "round them up or round them down." For instance {2|||1||0|0} has a count of 1.25. Black to play can "round up" to 2, gaining 0.75 on average. White to play can play to {1||0|0}, which also gains 0.75, moving to a position with a count of 0.5. But then Black can play to 1. The second player can always restrict the local result to the nearest integer. That is how actual fractions work, and so in chilled go we can treat such positions as actual fractions, with the ko caveat. |
Author: | Gérard TAILLE [ Tue Nov 03, 2020 10:57 am ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Gérard TAILLE wrote: I progress in CGT but I have a basic question. We found a number of examples of uncomparable positions. My question is the following : let's take two positions P1 and P2, each one having a miai value less or equal to 1. Can P1 and P2 be uncomparable? If yes I am interested by an example If no I am interested by a clue for the proof Positions where the top plays have miai value of can indeed be incomparable. For instance, {2|0} and {3|1||0}. See https://senseis.xmp.net/?GoInfinitesimals . Also, {3|3} is incomparable with 3, strictly speaking. ![]() However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Chilling is derived there, but the derivation uses aspects of CGT with which I am not familiar. FWIW, it seems to me that chilling works mainly because scores in regular go are numbers. Also, because the values of positions with miai values less than 1 are bounded by the nearest integers. As David Wolfe said, you can "round them up or round them down." For instance {2|||1||0|0} has a count of 1.25. Black to play can "round up" to 2, gaining 0.75 on average. White to play can play to {1||0|0}, which also gains 0.75, moving to a position with a count of 0.5. But then Black can play to 1. The second player can always restrict the local result to the nearest integer. That is how actual fractions work, and so in chilled go we can treat such positions as actual fractions, with the ko caveat. Yes Bill the answer to my question allows to justify chilling. Because chilling has been defined in "Mathematical Go" I suspect the answer to my question is "no" but we have to avoid a loop. The answer to my question allows to use chilling but you cannot use chilling to answer my question. Let's try maybe a preliminary question. Take a game G with a miai value equal to "m" and a score equal to "s". More generally, is it true that the result of the game when black (resp. white) plays first is a number between s+m (resp s-m) and s ? |
Author: | Bill Spight [ Tue Nov 03, 2020 1:18 pm ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Bill Spight wrote: Gérard TAILLE wrote: I progress in CGT but I have a basic question. We found a number of examples of uncomparable positions. My question is the following : let's take two positions P1 and P2, each one having a miai value less or equal to 1. Can P1 and P2 be uncomparable? If yes I am interested by an example If no I am interested by a clue for the proof Positions where the top plays have miai value of can indeed be incomparable. For instance, {2|0} and {3|1||0}. See https://senseis.xmp.net/?GoInfinitesimals . Also, {3|3} is incomparable with 3, strictly speaking. ![]() However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Chilling is derived there, but the derivation uses aspects of CGT with which I am not familiar. FWIW, it seems to me that chilling works mainly because scores in regular go are numbers. Also, because the values of positions with miai values less than 1 are bounded by the nearest integers. As David Wolfe said, you can "round them up or round them down." For instance {2|||1||0|0} has a count of 1.25. Black to play can "round up" to 2, gaining 0.75 on average. White to play can play to {1||0|0}, which also gains 0.75, moving to a position with a count of 0.5. But then Black can play to 1. The second player can always restrict the local result to the nearest integer. That is how actual fractions work, and so in chilled go we can treat such positions as actual fractions, with the ko caveat. Yes Bill the answer to my question allows to justify chilling. Because chilling has been defined in "Mathematical Go" I suspect the answer to my question is "no" but we have to avoid a loop. The answer to my question allows to use chilling but you cannot use chilling to answer my question. The two go together. Mathematical Go justifies chilling by other means. Chilling says that a position worth, say, 12.25 with an unchilled miai value of 0.75 is less than a position worth 12.5 with an unchilled miai value of 0.5, and that also means that playing in the 12.25 point position dominates playing in the 12.5 point position, subject to the ko caveat, OC. ![]() Quote: Let's try maybe a preliminary question. Take a game G with a miai value equal to "m" and a score equal to "s". More generally, is it true that the result of the game when black (resp. white) plays first is a number between s+m (resp s-m) and s ? Yes. The results of minimax play in G at temperature 0 are called the stops of G, and they will lie between s+m and s-m. ![]() |
Author: | Gérard TAILLE [ Tue Nov 03, 2020 1:27 pm ] |
Post subject: | Re: Thermography |
Bill Spight wrote: However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Let's try a position with t1 = 1 and t2 = 0.5 Are these two positions comparable? Let's build the difference game: That means that the two positions are not comparable, are they? |
Author: | Bill Spight [ Tue Nov 03, 2020 1:38 pm ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Bill Spight wrote: However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Let's try a position with t1 = 1 and t2 = 0.5 Are these two positions comparable? Let's build the difference game: That means that the two positions are not comparable, are they? True. ![]() However, the play with the miai value of 1 dominates the play with the miai value of ½. ![]() |
Author: | Gérard TAILLE [ Tue Nov 03, 2020 1:43 pm ] |
Post subject: | Re: Thermography |
Bill Spight wrote: However, the play with the miai value of 1 dominates the play with the miai value of ½. ![]() I understood that if two positions are uncomparable then, by definition, none of them could dominates the other. How do you define "dominate" ? |
Author: | Bill Spight [ Tue Nov 03, 2020 1:56 pm ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Bill Spight wrote: However, the play with the miai value of 1 dominates the play with the miai value of ½. ![]() I understood that if two positions are uncomparable then, by definition, none of them could dominates the other. How do you define "dominate" ? With the ko caveat, if one play is always at least as good as another one, it dominates it. If two plays are equally good, then they dominate each other. Positions (combinatorial games) do not dominate one another. They are either comparable, just as numbers are, or they are incomparable (confused with each other). Both positions and plays may be incomparable. ![]() |
Author: | Bill Spight [ Tue Nov 03, 2020 2:06 pm ] |
Post subject: | Re: Thermography |
Gérard TAILLE wrote: Bill Spight wrote: However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Let's try a position with t1 = 1 and t2 = 0.5 Are these two positions comparable? You have shown that they are confused in regular go, but, as you know, they are comparable in chilled go. The position on the left in chilled go is {1|1} and the one or the right is 0.5, which is less than {1|1}. |
Author: | Gérard TAILLE [ Tue Nov 03, 2020 3:15 pm ] |
Post subject: | Re: Thermography |
Bill Spight wrote: Gérard TAILLE wrote: Bill Spight wrote: However, if Move1 has a miai value, t1 <= 1, and Move2 has a miai value, t2 < t1, then Move1 dominates Move2, with the ko caveat. This is shown in "Mathematical Go" via the concept of chilling (See https://senseis.xmp.net/?Chilling ). Let's try a position with t1 = 1 and t2 = 0.5 Are these two positions comparable? You have shown that they are confused in regular go, but, as you know, they are comparable in chilled go. The position on the left in chilled go is {1|1} and the one or the right is 0.5, which is less than {1|1}. Isn't it obvious Bill? Take two positions confused in regular go with miai values t1≥t2. These positions becomes comparable under a tax≥t1 don't they? In the particular case of t1≤1 then the two positions are comparable in chilled go (tax=1). |
Page 9 of 19 | All times are UTC - 8 hours [ DST ] |
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |