Chess

[Gray Defender]

This is much better in all ways.  It was difficult to get a good image due to flickering visible on camera but I thought this one came out nice.  From the Atari 2600 Jr. Composite...

 

[+Andrew Davie]

4 hours ago, Gray Defender said:

I was able to load it with my Unocart, although it does make some atarivox babble when the program starts up.  I attached what it looks like on my crt.  Not as much contrast as I would like, but I could try to adjust the TV settings...

Thanks for the picture. The game relies on colour saturation rather than contrast, really. To me this picture looks quite washed out and too bright. I'd be turning down the brightness, and increasing colour saturation if you have those controls.

I did say the AtariVox babbled . -- it's quite fascinating, actually, right on the edge of something you think you should be able to understand.

 

[+Andrew Davie]

38 minutes ago, Gray Defender said:

This is much better in all ways.  It was difficult to get a good image due to flickering visible on camera but I thought this one came out nice.  From the Atari 2600 Jr. Composite...

 

 

Gorgeous. The flickering is from the TV and interaction with the camera/phone update rate, and possibly even with fluorescent lights if you have any of those - the game itself has zero flickering going on.  But, it's a lovely picture.  On a suitable TV it sure looks nice.

 

[Voxel]

Spent some hours trying a game against Video Chess.  So as not to be accused of any bias against Video Chess, it got to go first as white and it got to play on Game 5, so it had a clear advantage.  This is how the game played out...

 

Move	Video Chess	Game 5		AD Chess	chess20200419_3ply_quiescent#2_5ply
	White			Black
1	Pawn	D2-D4		Pawn	E7-E6
2	Knight	G8-F3		Bishop	F8-B4
3	Knight	B2-C3		Knight	B8-C6
4	Pawn	E2-E4		Bishop	B4-A5
5	Bishop	F1-C4		Bishop	A5-B4
6	Bishop	C1-F4		Bishop	B4-A5
7	King	E1-G1	Castle (Rook H1-F1)	Knight	G8-F6
8	Knight	F3-E5		Bishop	A5-C3	Knight capture
9	Pawn	B2-C3	Bishop capture	Knight	F6-E4	Pawn capture
10	Queen	D1-F3		Queen	D8-H4
11	Knight	E5-C6	Knight capture	Pawn	D7-C6	Knight capture
12	Queen	F3-E4	Knight capture	Bishop	C8-D7
13	Queen	E4-E5		Pawn	B7-B5
14	Queen	E5-G7	Pawn capture	Rook	H8-F8
15	Bishop	F4-H6		Queen	H4-E7
16	Queen	G7-F8	Capture Rook/Check King E8	Queen	E7-F8	Queen capture
17	Bishop	H6-F8	Queen capture	King	E8-F8	Bishop capture
18	Bishop	C4-B3		Rook	A8-D8
19	Rook	F1-E1		King	F8-G7
20	Rook	A1-B1		Pawn	F7-F5
21	Bishop	B3-E6	Pawn capture	Pawn	F5-F4
22	Bishop	E6-B3		Rook	D8-E8
23	King	G1-F1		Rook	E8-D8
24	Rook	E1-E7	Check King G7	King	G7-F8	checked
25	Rook	E7-F7		Pawn	A7-A5
26	Rook	F7-F4	Pawn capture	Pawn	B5-B4
27	Rook	F4-F3		Pawn	B4-C3	Pawn capture
28	Rook	F3-C3	Pawn capture	Bishop	D7-F5
29	Rook	C3-C6	Pawn capture	Rook	D8-D4	Pawn capture
30	Rook	C6-C7	Pawn capture	Rook	D4-D2
31	Rook	C7-C5		Bishop	F5-G6
32	Pawn	F2-F3		Pawn	A5-A4
33	Bishop	B3-A4	Pawn capture	Bishop	G6-F7
34	Pawn	A2-A3		Bishop	F7-D5
35	Pawn	C2-C4		Bishop	D5-F7
36	Pawn	F3-F4		King	H8-G8
37	Pawn	F4-F5		Bishop	F7-H5
38	Pawn	F5-F6		Bishop	H5-E2	check King F1
39	King	F1-E1	Checked	Rook	D2-A2
40	Rook	C5-C7		Bishop	E2-G4
41	Rook	B2-B8	Mate King G8

 

Edited April 18, 2020 by Voxel

[Mr SQL]

1 hour ago, Voxel said:

Spent some hours trying a game against Video Chess.  So as not to be accused of any bias against Video Chess, it got to go first as white and it got to play on Game 5, so it had a clear advantage.  This is how the game played out...

Nice breakdown! I think the 4-ply may do better, level 5 is gives Video Chess 10-12 minutes to move, I think it goes deeper than 3 moves ahead on 5.

 

[+Andrew Davie]

2 hours ago, Voxel said:

Spent some hours trying a game against Video Chess.  So as not to be accused of any bias against Video Chess, it got to go first as white and it got to play on Game 5, so it had a clear advantage.  This is how the game played out...

 

Move	Video Chess	Game 5		AD Chess	chess20200419_3ply_quiescent#2_5ply
	White			Black
1	Pawn	D2-D4		Pawn	E7-E6
2	Knight	G8-F3		Bishop	F8-B4

 

 

Thanks for that!  To be clear, you played a level that is allowed 10-12 minutes per move? Video Chess is quite the accomplishment.

I'm going to have to put some smarts in that evaluation algorithm sometime and see how that spices things up. But first I'm going to get all the other mechanics as good as I can. Then it will be time to put some "thinking" in.

[NE146]

Re: thinking time on the VCS, would it affect either game if they were run in Stella for example, and increasing the emulation speed under video options to 1000%? i.e. To decrease waiting time?

 

 

[+Andrew Davie]

Just now, NE146 said:

Re: thinking time on the VCS, would it affect either game if they were run in Stella for example, and increasing the emulation speed under video options to 1000%? i.e. To decrease waiting time?

Nice idea, but the UI becomes totally unusable for my game when run at super-speed. I can't see an on-the-fly way to change stella emulation speed, so unless you could do this, then I don't think this would work.

[+fdr4prez]

24 minutes ago, NE146 said:

Re: thinking time on the VCS, would it affect either game if they were run in Stella for example, and increasing the emulation speed under video options to 1000%? i.e. To decrease waiting time?

 

 

 

Stephena says it will speed everything up

 

curious if someone would give it a try

 

He says that the input would also be increased, so it may make it harder to move.

 

[+Andrew Davie]

Have you ever seen the movie "The Incredibles" where the boy was in a foot-race and he was trying to win but not win so incredibly well that people would notice he's a superhero?  Thats what my chess program has been doing. More to the point, it tries very hard to beat the opponent, without ACTUALLY beating the opponent. It will make every effort to win but NEVER take an opportunity to win.

Ha!  Bizarre, but a consequence of the recent code I put in to handle check, and backing out so the previous ply would know it was an illegal move. Whoops!  I guess this kind of invalidates the Video Chess comparison - but not really. The comparison was against the engine in its current state, bugs and all.  But anyway, I need to find a way to handle check that is compatible with the negamax and quiescent code.

Meanwhile, I now have illegal moves no longer available for the player. Or, mostly. I've been holding off on this until I found an efficient way to handle it. It's done by a scan, just after all moves are generated, so see if any of the capture moves are capturing a king. Then I set a flag. This is fairly efficient because the move list is sorted, with capture moves first. And now that's in/working, I can extend it to work for castling. Basically, the king when castling will "cross" two squares. I'll have two vars (king1, king2) which are normally 0, but when castling, these are set to the squares the king crosses. And the move gen code which scans for capture of a king can see when those vars are non-0 and do a scan for any moves with those as destination squares. If so, then just like being in-check, a flag is set to indicate the previous move was illegal.

This is a relatively efficient way to do things, I think.

Meanwhile, I'm under pressure to find a better way to handle the check stuff in-game. I've disabled it now and it readily performs checkmates against me in test scenarios. When I gave it two queens and a king against my king before, it found it impossible to win. Now it readily either checkmates or stalemates me. I'm really just dealing with fringe/edge cases. A stalemate is effectively "take the king on the next move" sort of thing. I have a bit of tinkering to do. Frustrating.

[Voxel]

8 hours ago, Andrew Davie said:

To be clear, you played a level that is allowed 10-12 minutes per move?

Before I started the game to confirm what I was doing I referred to the manual.  Game 5 is 3minutes 15 seconds, which I deemed more than satisfactory.

8 hours ago, Andrew Davie said:

I'm going to have to put some smarts in that evaluation algorithm sometime and see how that spices things up. But first I'm going to get all the other mechanics as good as I can. Then it will be time to put some "thinking" in.

Please can I make a request.  When the engine has finished it's rather nice animated move, can the cursor be positioned from where the piece moved and that the moved piece flashes.  I could easily miss which piece has moved when pitting two engines together, if I happen to get distracted during the movement made.

 

The game was very much evident as between two different game styles.

 

I'd like to add that Video Chess played a very by the book game.  I have no idea what chess knowledge is in it but it plays very safe calculated moves.  It had one clear advantage seen in the early stages of the game where it knew to castle early on, develop its' pieces and connect the rooks.  That's a text book opening.  This early development served it well later in the game against the version of chess played.  I should add that Video Chess was not quick in making moves (even move 17 - Queen capture was thought about).

 

As for Chess (chess20200419_3ply_quiescent#2_5ply), it played a very direct game.  Moves 5 (Bishop) and move 10 (Queen), show how this engine thinks.  When you see those diagonal moves from these pieces it has a plan.  I found this current engine now mobilises minor pieces like Bishops and Knights first.  It has an inclination to sacrifice Bishops for Knights, and is very happy to exchange pieces.  It still moves the queen early and tends to save the Rooks for later.  I understand this is still at alpha, not having all functions implemented so it's game can only get better.

 

I shall look forward to this game putting on its' super hero suite, as there are other games I would like to play this against.

[+Andrew Davie]

2 hours ago, Voxel said:

Please can I make a request.  When the engine has finished it's rather nice animated move, can the cursor be positioned from where the piece moved and that the moved piece flashes.  I could easily miss which piece has moved when pitting two engines together, if I happen to get distracted during the movement made.

That's an excellent suggestion. I, too, have missed which piece has moved regularly. I've made the change in this new version. There's a double-flash after the computer makes its move, which I hope to get rid of soon. But you can see what you have requested is implemented, and I think it works quite well. Thanks.
 

 

 

 

chess20200419_UI.bin

[+Andrew Davie]

I really like that from/to indication. I've put the same thing in for the player, too. Seems to work.

Here's a minor update with that change, and a longer gameplay video, just because I can.
This is getting close to the version I'll "submit" for the @ZeroPage Homebrew show. I don't think I'm going to be able to fix up all of the issues with "in check" and what to do there. The user's move selection, though, is now correct - except for castling in check - so that's a plus.

 

 

Things to note in the video -- the "from" square is now highlighted at the start of any move.

When castling, both the king move and the rook move have the start square highlighted but as the rook moves last, it is the rook that does the flashing.
 

chess20200419_UI2.bin

Edited April 19, 2020 by Andrew Davie

[Voxel]

Like a Christmas wish - thanks!

[Voxel]

I'd like to enquire about your thoughts on the implementation of best line played last go, being carried over as a strong candidate in the search for the next move.

5 hours ago, Andrew Davie said:

Have you ever seen the movie "The Incredibles" where the boy was in a foot-race and he was trying to win but not win so incredibly well that people would notice he's a superhero?

I'm following on from your line above, but having the engine think of a worse case scenario to it's moves and be prepared to play them.

 

So engine searches, and says I'm going to play this great move. Engine makes move.  Engine also looks ahead and sees humans' next possible best moves in reponse and says here are a few.  I'll remember them as possibilities (only the worst case scenarios it might need to deal with).  Human moves.  Engine then picks up the stored tree with moves and says, human went with that move, I'll see if it matches any of the ones stored.  If it matches the engine searches the next series along that tree in addition to any others in its' algorithm for other pieces on the board for its' next move candidate. It would see when the tree is exhausted and null it so as to not persue it further.

[+Karl G]

I like the new move indicator quite a bit. Since you are enjoying pictures of the game on CRT, here is mine running on my modded Jr hooked up to a tiny 9" Sony Triniton TV. 

 

[+Andrew Davie]

29 minutes ago, Karl G said:

I like the new move indicator quite a bit. Since you are enjoying pictures of the game on CRT, here is mine running on my modded Jr hooked up to a tiny 9" Sony Triniton TV. 

 

Thanks for that. Beautiful. I think the tiny colour TVs just work the best for this sort of graphics.

Here's an interesting aside. By now you're familiar with the pieces and they probably look all pretty easy to recognise.

Turn your TV (or the picture!) upside-down and see how much, much more difficult they are to recognise.

 

[+Karl G]

6 minutes ago, Andrew Davie said:

Turn your TV (or the picture!) upside-down and see how much, much more difficult they are to recognise.

Huh. It didn't that much more difficult to me. I guess maybe my brain's pattern recognition software is slightly different.

[Mr SQL]

Just played against the latest engine - it is moving really fast is this the 4ply version? The UI is really smooth, it did well at first then seemed to be moving to fast/not going enough moves deep to be 4x I think.

 

I took a pic on my cell - have to figure out how to get it off of there again now - it looks like I have brighter artifact colors than Carl and Gray Defender, likely due to RF and no composite mod.

 

Here is a breakdown pic from earlier discussion on the different atrifact colors available over NTSC from the patterns - I didn't find the crumpled paper pic with the 8 artifact colors but this one has 4 or 5, the emu is not showing them all I think:

[+Andrew Davie]

Here's a version where I've tweaked the "position" tables a bit, the effect of which appears to improve the "urgency" for the computer to develop pieces and more particularly castle early, and indirectly connect the rooks. It's amazing what tweaking a few numbers can do to change the gameplay. These tables will undergo lots of refinement, but I thought I'd share the result of the first pass...

 

chess20200420.bin

[+Andrew Davie]

10 hours ago, Mr SQL said:

I took a pic on my cell - have to figure out how to get it off of there again now

Two easy options - email to yourself, if you have email setup on your phone.

Or, install Dropbox on phone and your laptop, then just drop it in your Dropbox folder.

[+Andrew Davie]

Here is my first candidate version for playing/demo on the @ZeroPage Homebrew show this Wednesday.

It has a whopping 11-ply quiescence search (!!) although in testing I've only seen it go to 11-ply

The actual search is a mere 3-ply, so it's generally going to play moves quite quickly.

 

1) no en-passant moves

2) you can castle in/across check

3) no repetitive draw - consider if you get repetition YOU have to choose a different move (although the computer does have SOME randomness)

4) when all is lost, the computer accepts loss of its king and fights on. If you take the computer king before he takes yours, you win

 

Any feedback gratefully accepted!

 

 

 

chess20200420_ZPH_Candidate1.bin

Edited April 20, 2020 by Andrew Davie

[+Andrew Davie]

On 4/19/2020 at 7:13 AM, Gray Defender said:

This is much better in all ways.  It was difficult to get a good image due to flickering visible on camera but I thought this one came out nice.  From the Atari 2600 Jr. Composite...

Here's a tip you can try to reduce flicker when taking pictures...

TV in front of a wall, preferably light in colour.

First, put a light/lamp behind your back, pointing "towards" the TV, but not directly at it.  The lamp highilghts the wall behind the TV. Rest of the lights in the room off. So, only the lamp giving illumination in the room. The camera sees the light reflected off the wall, and the image on the TV. You want to be able to shift your body to obscure some of the light heading towards the wall.  Now, as you point your cam to the TV, the exposure rate will automatically adjust, based on the light it detects. You want to now shift your body to change the background light the camera detects, so it increases or decreases its exposure time. You want the exposure timing to match a multiple of the frame rate. You should see this by reduction in flicker on the camera view/image. It sounds screwy, but actually works - at least for my mobile phones.  I figured this out when trying to take images of my mechanical TV's display, which has a frame rate of 12.5Hz.

 

Edited April 20, 2020 by Andrew Davie

[+Andrew Davie]

17 hours ago, Mr SQL said:

Just played against the latest engine - it is moving really fast is this the 4ply version? The UI is really smooth, it did well at first then seemed to be moving to fast/not going enough moves deep to be 4x I think.

 

If it's moving fast, almost certainly 3-ply.

I plan to dynamically change the depth searched, based on the # nodes visited in the previous move cycle.

[+Andrew Davie]

18 hours ago, Voxel said:

I'd like to enquire about your thoughts on the implementation of best line played last go, being carried over as a strong candidate in the search for the next move.

I'm following on from your line above, but having the engine think of a worse case scenario to it's moves and be prepared to play them.

 

So engine searches, and says I'm going to play this great move. Engine makes move.  Engine also looks ahead and sees humans' next possible best moves in reponse and says here are a few.  I'll remember them as possibilities (only the worst case scenarios it might need to deal with).  Human moves.  Engine then picks up the stored tree with moves and says, human went with that move, I'll see if it matches any of the ones stored.  If it matches the engine searches the next series along that tree in addition to any others in its' algorithm for other pieces on the board for its' next move candidate. It would see when the tree is exhausted and null it so as to not persue it further.

 

Your thinking seems sound to me, but don't forget this is running on a very underpowered Atari 2600.  Even though I have extra ROM/RAM to play with, these come with significant penalties/limitations, particularly with this bank-switching scheme (3E).  What you are describing are the sorts of things that would make a lot of sense on a platform that has some decent memory and processing power
I don't think it's going to happen on this little effort, I'm afraid!  I'm really, really struggling to get anything new put in at the moment, due to being out of ROM space in the banks that I really need it in.