Final Fantasy VIII

2012-06-27 06:48:55 pm

Moo! Flap! Hug!

I've posted a new version of my Triple Triad assistant. It's a beta at the moment, and I'd love any feedback on it. In particular, it now includes typo-protection (see the bottom of the Card class), and I hope there are no bugs in there. It also includes a less-verbose mode so that there's less clutter on the screen during a run (if you preferred the old way, you can just set it back on in TriadHelper).

I've also factored out all user input and output into the IOMethod interface, of which only Console is currently implemented. I have something interesting in mind on that front Wink

2012-06-27 11:51:32 pm

SDA Speedruns: 1

Its so buggy, dont use it.

2012-06-27 11:58:05 pm

Moo! Flap! Hug!

New version posted (1.1-beta2). FFgamer86 was right, in a sense. We haven't found any bugs in the code itself. But, the "anti-typo" checking worked by comparing your input against a listing of the 110 cards available in the game, to make sure that your input matched one of the cards. Unfortunately, I automatically generated that list of the 110 cards by parsing an FAQ on GameFAQs...enough said. Basically, the list was wrong beyond all belief (it got card values wrong left and right...and even up and down).

A huge thank you to FFgamer86 who went through the 110 entries in my program with me over Skype and helped me correct them all. If a third party wants to double-check the list, it's at the very bottom of Card.java, and it should be readable even to a non-programmer.

2012-06-28 12:01:21 am

SDA Speedruns: 1

Also, all the cards are read top right bottom left. So Ifrit would be read 9 6 2 8 and looks like this http://www.ffextreme.com/site/ff8/cards/8-8.jpg

Edit history:

Poxnor: 2012-06-29 04:00:18 am

2012-06-29 01:49:50 am

Moo! Flap! Hug!

Caught an actual bug in the code Sad

1.1-beta3 posted.

2012-06-29 04:01:00 am

Moo! Flap! Hug!

Err, sorry, posted in this one (I posted the wrong file to the previous post, and I couldn't figure out how to edit in the attachment).

TriadHelper - Executable.zip

2012-07-02 09:42:18 pm

It took all of about 30 minutes to port your code over to C#. It works perfectly fine as a console program and I've attached both the source and executable. The layers are somewhat nonstandard in the .NET world, so I'm going to change the code around before I do anything else with it. Also I've attached a preview of what I'm planning.

Attachments:

TriadHelper - Source.zip

ttapreview.png

Carcinogen

2012-07-04 05:11:58 am

Very interesting, parsing from 3 sides instead of four.

Edit history:

FFgamer86: 2012-07-04 03:17:14 pm

2012-07-04 03:10:46 pm

SDA Speedruns: 1

Also of note, The wheel that is under balamb garden that you mash square to turn. If you fail on it with all 3 characters. You get to do it over again with all 3 characters. Also with all 3 characters, you only have to push square twice to succeed in turning the wheel. I am testing the 2 person one now to see how many presses for that. I won't be able to figure out the squall only one though. If someone with an emulator wants to figure it out, would be helpful.

Edit: It seems after you fail with 3 people. The 4th time is unfailable. The 3 person one is not 2 presses to get.

2012-07-04 09:32:18 pm

Actually it parses from as many sides as it takes to uniquely identify the card. i.e. Caterchipillar only requires you to type "42" but Creeps requires you to type "5252".

I actually finished up most of the functionality of the program now, but I'm a little disappointed with the performance. Especially since I'm having it analyze two sides at once for the first move (the longest move to calculate, of course.) I plan to take a profiler to it to see if there's a bottleneck somewhere, since pretty much what I did is copy paste the java code and make adjustments until it compiled. Lucky when I compiled it the first time, it gave the appropriate output.

2012-07-13 05:08:05 am

Moo! Flap! Hug!

Quote from Dessyreqt:

I actually finished up most of the functionality of the program now, but I'm a little disappointed with the performance. Especially since I'm having it analyze two sides at once for the first move (the longest move to calculate, of course.)

I'm not sure what you mean by analyzing two sides at once for the first move. For performance comparison, the code I posted can analyze the first move of the game in about 3 seconds on my laptop for an average game.

Edit history:

Poxnor: 2012-07-13 06:30:55 am

Poxnor: 2012-07-13 06:30:46 am

2012-07-13 06:20:29 am

Moo! Flap! Hug!

Quick note: some time back in this thread, I posted an analysis of 131072 hands played by each of the three Trepies, as well as Zell's mom. The little program that analyzed the dump of 131072 hands was using the same buggy listing of cards that my Triple Triad helper was using back in 1.1-beta1 (incorrect values on card edges, etc.). Here are some corrected figures.

Trepie #1:

Code:

Total hands processed:         131072
Unique unordered hands played: 26100
Unique cards played:           22
Hands with top-level card:     39339 (30.01%)
Average card strength:         4.48
Levels of cards played:        2, 5, 10

Trepie #2:

Code:

Total hands processed:         131072
Unique unordered hands played: 100528
Unique cards played:           33
Hands with top-level card:     26227 (20.01%)
Average card strength:         4.13
Levels of cards played:        1, 3, 5, 10

Trepie #3:

Code:

Total hands processed:         131072
Unique unordered hands played: 123543
Unique cards played:           44
Hands with top-level card:     13111 (10.00%)
Average card strength:         4.07
Levels of cards played:        1, 2, 4, 5, 10

Zell's Mom:

Code:

Total hands processed:         131072
Unique unordered hands played: 123393
Unique cards played:           44
Hands with top-level card:     13116 (10.01%)
Average card strength:         4.08
Levels of cards played:        1, 2, 4, 5, 10

Each of the above will never play level 5 card Pupu #47, but will play one level 10 card (Quistis #103 or Zell #105). Hence, the number of unique cards they play is a nice multiple of one less than the number of levels of cards they can play.

Quick note: some time back in this thread, I posted an analysis of 131072 hands played by each of the three Trepies, as well as Zell's mom.  The little program that analyzed the dump of 131072 hands was using the same buggy listing of cards that my Triple Triad helper was using back in 1.1-beta1 (incorrect values on card edges, etc.).  Here are some corrected figures.

Trepie #1:
[code]

Total hands processed:         131072
Unique unordered hands played: 26100
Unique cards played:           22
Hands with top-level card:     39339 (30.01%)
Average card strength:         4.48
Levels of cards played:        2, 5, 10
[/code]
Trepie #2:
[code]

Total hands processed:         131072
Unique unordered hands played: 100528
Unique cards played:           33
Hands with top-level card:     26227 (20.01%)
Average card strength:         4.13
Levels of cards played:        1, 3, 5, 10
[/code]
Trepie #3:
[code]

Total hands processed:         131072
Unique unordered hands played: 123543
Unique cards played:           44
Hands with top-level card:     13111 (10.00%)
Average card strength:         4.07
Levels of cards played:        1, 2, 4, 5, 10
[/code]
Zell's Mom:
[code]

Total hands processed:         131072
Unique unordered hands played: 123393
Unique cards played:           44
Hands with top-level card:     13116 (10.01%)
Average card strength:         4.08
Levels of cards played:        1, 2, 4, 5, 10
[/code]
Each of the above will never play level 5 card Pupu #47, but will play one level 10 card (Quistis #103 or Zell #105).  Hence, the number of unique cards they play is a nice multiple of one less than the number of levels of cards they can play.

Edit history:

Poxnor: 2012-07-13 12:38:45 pm

Poxnor: 2012-07-13 12:38:22 pm

Poxnor: 2012-07-13 12:36:10 pm

2012-07-13 12:25:50 pm

Moo! Flap! Hug!

New version of the Triad Helper posted. I'd appreciate any feedback, because I'm going to be using this in a charity marathon (TSG Community FF marathon supporting To Write Love On Her Arms) in a little over a week. I'm calling it 1.9-beta4 (betas 1-3 of 1.9 have been internal between FFgamer and myself), with the intention that this will ultimately be version 2 once finalized.

New features:
- Keeps track of cards, tells you which card to take (so you're not constantly re-inputting your own cards at each new game);
- Can have opponent concede or abort a game that you can see how to win/draw respectively (allowing you to return to the card-tracking menu); and,
- Audio support, i.e., voice input and speech output (OS X only for now...and as long as I'm the sole developer of this branch, lol). NB: the underlying Sphinx-4 library for voice input has a heck of a time recognizing the word "five" (at least with my accent), so the program alternately accepts the word "penta" for anywhere you could say "five," e.g., "play penta one three penta to spot penta." The grammar isn't documented in the README yet, but I'm assuming that anyone interesting in beta-testing this program could read the tripleTriad.gram grammar file Smiley

Read the README file regarding compilation (seriously; if you try to compile with audio support without building the required library -- which I cannot redistribute for licensing reasons -- the entire mess will fail. Build without audio support using "ant noaudio"). Sorry, questions about using this version in Windows will have to be answered by someone other than me; ask away about OS X / UNIX support though.

If you try out bin/audioHelper (the audio-support version for OS X), note the "-careful" and "-rude" flags. The "-careful" flag is useful for marathons, since it will confirm what it hears from you. The "-rude" flag is a donation incentive, lol, but it's still pretty tame (suggestions appreciated!).

Basically, this program is now becoming what it should have been for SGDQ (my sincere apologies to SDA and OAR for not having this ready in-time).

New version of the Triad Helper posted.  I'd appreciate any feedback, because I'm going to be using this in a charity marathon (TSG Community FF marathon supporting To Write Love On Her Arms) in a little over a week.  I'm calling it 1.9-beta4 (betas 1-3 of 1.9 have been internal between FFgamer and myself), with the intention that this will ultimately be version 2 once finalized.

New features:
- Keeps track of cards, tells you which card to take (so you're not constantly re-inputting your own cards at each new game);
- Can have opponent concede or abort a game that you can see how to win/draw respectively (allowing you to return to the card-tracking menu); and,
- Audio support, i.e., voice input and speech output (OS X only for now...and as long as I'm the sole developer of this branch, lol).  NB: the underlying Sphinx-4 library for voice input has a heck of a time recognizing the word "five" (at least with my accent), so the program alternately accepts the word "penta" for anywhere you could say "five," e.g., "play penta one three penta to spot penta."  The grammar isn't documented in the README yet, but I'm assuming that anyone interesting in beta-testing this program could read the tripleTriad.gram grammar file :)

Read the README file regarding compilation (seriously; if you try to compile with audio support without building the required library -- which I cannot redistribute for licensing reasons -- the entire mess will fail.  Build without audio support using "ant noaudio").  Sorry, questions about using this version in Windows will have to be answered by someone other than me; ask away about OS X / UNIX support though.

If you try out bin/audioHelper (the audio-support version for OS X), note the "-careful" and "-rude" flags.  The "-careful" flag is useful for marathons, since it will confirm what it hears from you.  The "-rude" flag is a donation incentive, lol, but it's still pretty tame (suggestions appreciated!).

Basically, this program is now becoming what it [i]should[/i] have been for SGDQ (my sincere apologies to SDA and OAR for not having this ready in-time).

2012-07-14 02:21:55 pm

Quote from Poxnor:

Quote from Dessyreqt:

I actually finished up most of the functionality of the program now, but I'm a little disappointed with the performance. Especially since I'm having it analyze two sides at once for the first move (the longest move to calculate, of course.)

I'm not sure what you mean by analyzing two sides at once for the first move. For performance comparison, the code I posted can analyze the first move of the game in about 3 seconds on my laptop for an average game.

Well, I wrote with a slightly different use case in mind. I was using an online triple triad flash game AI to test my version of the bot, and on that game you can start as either blue or red. I've also implemented Same, Same Wall, Plus, and Combo. I haven't been able to figure out a good way to implement Closed, and I've been too lazy to implement Elemental. However the first move for each side can take up four times as long as your version. I'm pretty sure this is because I didn't really make any optimizations to the code, I just copied your code and fixed all the syntax errors. I've added you on Skype if you'd like to discuss this further, though I'm not usually on there, I'm usually just in IRC on the SDA server.

[quote="Poxnor"][quote="Dessyreqt"]I actually finished up most of the functionality of the program now, but I'm a little disappointed with the performance. Especially since I'm having it analyze two sides at once for the first move (the longest move to calculate, of course.)[/quote]
I'm not sure what you mean by analyzing two sides at once for the first move.  For performance comparison, the code I posted can analyze the first move of the game in about 3 seconds on my laptop for an average game.[/quote]
Well, I wrote with a slightly different use case in mind. I was using an online triple triad flash game AI to test my version of the bot, and on that game you can start as either blue or red. I've also implemented Same, Same Wall, Plus, and Combo. I haven't been able to figure out a good way to implement Closed, and I've been too lazy to implement Elemental. However the first move for each side can take up four times as long as your version. I'm pretty sure this is because I didn't really make any optimizations to the code, I just copied your code and fixed all the syntax errors. I've added you on Skype if you'd like to discuss this further, though I'm not usually on there, I'm usually just in IRC on the SDA server.

2012-07-14 03:20:14 pm

Moo! Flap! Hug!

Quote from Dessyreqt:

I haven't been able to figure out a good way to implement Closed

That's a thesis project.

Quote from Dessyreqt:

I've been too lazy to implement Elemental

Yeah, that's pretty boring. Let someone who really needs it implement that Wink

Quote from Dessyreqt:

However the first move for each side can take up four times as long as your version. I'm pretty sure this is because I didn't really make any optimizations to the code, I just copied your code and fixed all the syntax errors

That might have more to do with how the new rule-features were incorporated (??). But, for what it's worth, my code wasn't optimized at all. It was pretty textbook alpha-beta. So, I don't really know.

...But, I think that I should link this thread for the next time one of my students claims that Java is inherently slower than any C-based language, and that any Java program is always slower than any C-based port of said program Wink

2012-07-14 04:23:08 pm

Quote from Poxnor:

That might have more to do with how the new rule-features were incorporated (??). But, for what it's worth, my code wasn't optimized at all. It was pretty textbook alpha-beta. So, I don't really know.

...But, I think that I should link this thread for the next time one of my students claims that Java is inherently slower than any C-based language, and that any Java program is always slower than any C-based port of said program Wink

Well, the new features only come into play only when they are selected for use, basically I just added the logic to the board.play method and left the alphaBeta logic alone. I'm talking about the base Open-rule only. As far as comparing languages, I doubt C# is what your students are talking about when they say "C-based language" as C# is also compiled to an intermediate language which is then interpreted into Java. They're probably talking about C or C++, which compile directly to assembly readable by a target machine. In any case, from what I've read C# DOES generally perform better in most cases. But since neither your code nor my code have been optimized, it's not a good example either way.

[quote="Poxnor"]That might have more to do with how the new rule-features were incorporated (??).  But, for what it's worth, my code wasn't optimized at all.  It was pretty textbook alpha-beta.  So, I don't really know.

...But, I think that I should link this thread for the next time one of my students claims that Java is inherently slower than any C-based language, and that any Java program is always slower than any C-based port of said program ;)[/quote]

Well, the new features only come into play only when they are selected for use, basically I just added the logic to the board.play method and left the alphaBeta logic alone. I'm talking about the base Open-rule only. As far as comparing languages, I doubt C# is what your students are talking about when they say "C-based language" as C# is also compiled to an intermediate language which is then interpreted into Java. They're probably talking about C or C++, which compile directly to assembly readable by a target machine. In any case, from what I've read C# DOES generally perform better in most cases. But since neither your code nor my code have been optimized, it's not a good example either way.

zoonel

2012-07-15 03:59:55 am

Go play spacechem !

just to say that I’ve tested the java version on Ubuntu with open Jre and then on Windows xp with the sun jre and for the same hand with the same compiled stuff sun jre was 6 time faster.
Also the c# version speed is comparable to the java with sun jre on my computer.

2012-07-17 12:57:07 pm

Moo! Flap! Hug!

New version posted (1.9-beta5). Two key changes:

- There are times you actually want to take a really weak card from an opponent when you win (e.g., if the opponent is playing a higher-indexed card than the lowest-indexed card in your hand of five, but that higher-indexed card is weaker than your lowest-indexed card). There was an error in the logic of how to deal with these situations in 1.9-beta4; fixed here.

- A new "suspend program" command has been added to the audio program (press enter to resume); useful for commentating during marathons or streams.

I'm very happy with this version, and the speech recognition aspect is working beautifully now (it's faster and easier than typing inputs). This is probably pretty close to the finished version 2 that I'll put out soon. Suggestions for the "-rude" Siri are still welcome, as it'll be a donation incentive at the TSG Community marathon this Friday.

Caracarn

2012-07-19 12:14:20 pm

new RTA record 8:56:43 by トマトごはん (Tomato Rice). If you have the ability to watch nico live, here it is.

http://live.nicovideo.jp/watch/lv100892225?ref=community - Part 1

http://live.nicovideo.jp/watch/lv100905383?ref=community - Part 2

2012-07-22 05:19:29 am

Thanks for posting Caracarn