I had an idea for it. A couple years ago I did a bit of language processing in my Automata class. I was thinking of breaking things up that way and parsing each word of input to determine the action taken. Then once you know what the card is doing, you can evaluate x vs y on those actions. It seems elegant enough, but it's not the easiest thing in the world to write since I'm pretty bad at building those types of programs. Oracle text actually has a pretty good structure for parsing. I'm pretty sure Wizards did it that way intentionally.Originally Posted by Echelon
Reading up on fsm's and tokenized inputs has been on my to do list but life has been in the way, between graduating, moving, work, and so on.
My current plan, is that I have a bit of work travel coming up in a couple weeks and reading that type of very dry material seems like an excellent way to spend my time on the plane.
I tend to just watch movies when flying... Hurray for the iPad when there's no in-flight entertainment system.
Even that kind of language processing takes a lot of work and effort. You'll still manually have to verify whether the (sample) interpretations are correct or not. From what I've understood from a session I attended at/from one of our countries leading companies in this field is that the world of data science mostly consists of a little bit of cleverness and a whole lot of hard work. At the time they were working on software that could analyse x-ray images to detect tumors and other anomalies. Apparantly there's about a 40% error margin when analysed by human technicians (which shocked me tbh. If I ever get have to get any x-rays I'll be pretty damn sure to get a 2nd and 3rd opinion regardless of the outcome) so the motivation to get rid of human error was pretty damn big. They talked us through the process of how the software worked and came to be, which was very interesting.
I listened to a radio show about that xray analysis recently, it was pretty interesting.
My plan right now to error test is to take a small selection of cards and feed those in. I can do it manually and automatically and see if the results match. If a few well thought out test cases are accurate then I can be confident the rest of it works. But, given the rate at which this project is progressing I'm still probably a pretty long ways off from getting that done.
Most planes in the US have wifi these days, depending on the airline the cost varies but it's generally around $8-$10 for flight long wifi (in this case it's a 4 hour flight). It's not fast enough to stream video, but it's good for reading. Or I could just be a normal person and download the book I intend to read.
Yeah, just use a pool of a few simple cards and a couple of corner cases and keep expanding those corner cases as accuracy increases. Pretty much work your way down from broad/common to narrow/rare occurrences and more and more pieces start to fall into place as you go.
In a deck based on arena rector + PW, shalai, voice of plenty, how do you see it? is synergistic with the deck? since from hexproof to us, to creatures and planeswalkers.
Well, you can't GSZ for it and it doesn't play well w/ the Evolutionary Leap engine (would you choose to take that route). Furthermore the planeswalkers you do play (mostly Ugin/Elspeth) have a natural ability to rid themselves of any threats. That just leaves spell based planeswalker removal, which is relatively rare.
It's a good card, but probably not for this particular deck.
So a team of developers from Google makes an AI that plays Go, a simple game involving 2 players taking turns placing dots on a grid, and this makes international news.
Meanwhile some random forum user claims to have made an AI that plays MTG, and I'm the troll? Right.
Perhaps you might like to put your money where your mouth is and implement a bot that plays MTGO?
Ignoring the AI issue, I do believe you have created some kind of function that recommends you cards, but your inputs for this function are so limited that you don't get a better (or importantly, different) result than you would from:
a) Playtesting games yourself and using your intuition (assuming you're at least a decent player/deckbuilder to start with)
b) Exploring your ideas using pen and paper with a pros-cons list or some other kind of simple method for organizing your thoughts
If you want to work on this as a hobby for your own enjoyment then I have no problem with it and who knows, maybe one day you'll hit some breakthrough and really will have an AI that plays magic and builds decks, and I wish you the best of luck. But currently you (and everyone else) are nowhere close to this, so statements equivalent to "I play this card because the algorithm says so" aren't helpful for anyone and don't belong on message boards.
I have nothing more to say on this, so if you're confident that your system works then we'll just have to agree to disagree
ok, thank you very much for the explanation
Don't worry. I wouldn't waste any more time attempting to battle the elitist circle jerk that is this thread. Turns me off from even wanting to play the deck.
Plenty have already done this. DotP did this. There is a big difference between playing the game, which is quite trivial to implement, playing the game well, and playing the game perfectly (which is what the Go and Poker AI's in that past couple years were built to do).
Not true. And I can give you an example why. Magic has a very high degree of variance, so much so that even thousands of games tested will give you statistical results that it is quite obvious are incorrect. I'll even give you some actual results of some simple mana curves I did a couple months back to prove this point. This example tracked the number of lands drawn at 24, 25, and 26 landa) Playtesting games yourself and using your intuition (assuming you're at least a decent player/deckbuilder to start with)
https://imgur.com/a/ziQ1i
See the errors? The first one that should jump out at you is that 26 land had a 0 lander more often than 25 land did. And that was over 100,000 games.
People always go back and forth on land counts, they cut a land because they feel like they're flooding, or add a land because they were mana starved but the reality is that those types of situations play out over a handful of games, at the most 100 games.
When 100,000 games doesn't even give you a solid base to conclude that 24, 25, or 26 lands are correct, and considering that land drops are the most consistent action in almost all decks, then every conclusion you make about how a deck functions is also on equally shaky ground.
I've been testing the system for years now. The first variant played Modern Knightfall before that deck had any tournament success, and therefore no lists to base a deck skeleton off of. I was able to identify one, and that skeleton ended up getting extremely close to the deck in it's modern day form. After that I used Burn, and was able to identify cards that weren't being played that had potential. I've then added those cards and played it at tournaments, to decent success.If you want to work on this as a hobby for your own enjoyment then I have no problem with it and who knows, maybe one day you'll hit some breakthrough and really will have an AI that plays magic and builds decks, and I wish you the best of luck. But currently you (and everyone else) are nowhere close to this, so statements equivalent to "I play this card because the algorithm says so" aren't helpful for anyone and don't belong on message boards.
It's not perfect, but no one said it was. Echelon and I were just explaining to the person with their matrix that if you want to follow that approach, there's a much better way to do it which is with rule based evaluation card by card. Not a point based system for effects.
Edit: And all of this has an actual impact on discussing deck building choices. To give an example, something you'll see very often in Nic Fit is people playing a 61st card. This tends to go against almost all conventional wisdom for deck building. However, through various methods (logic/reasoning, a simulation such as my above example, a little bit of practice, and so on) we have over the years come to the conclusion that the toolbox value of having one additional card in the deck so that it's there when you need it, outweighs the inconsistency added to the deck. I've proven this in several decks now, so it's not just a toolbox related effect. Having an out to something unexpected that you can potentially draw in your deck as an additional card is much more valuable than adhering strictly to 60 cards, and that's something I've even started applying to other decks and formats.
Last edited by Brael; 06-16-2018 at 01:11 AM.
Have some examples? This is generally considered one of the friendliest threads on the forum from what I've seen.
The biggest issue a deck like Nic Fit has though, is that it attracts a lot of people new to the format and there's certain things that just don't work. For example, Pod is one of the weakest archetypes and BUG is the weakest color combination. Given the cost of the format, it's only fair to tell people up front if their build is or isn't viable.
Just one page ago, the almighty Echelon “yeah no you don’t know what you’re taking about.” Just rubs me the wrong way. Appreciate your thorough response, though. I’ve been lurking because the deck seems awesome. I understand legacy in general is cost prohibitive, but it’s a turn off for new members to get their ideas shot down in a condescending manner.
To pile onto Brael: in software development the scale of work generally isn't considered when talking about stuff like the core workings of an AI. Consider the AI that beat pro dota players last year. This sums up how it works: It's a neural network that loses points for taking damage and gains points for inflicting damage. Sounds simple, right? Also sounds like easy to make, right? It isn't though, it took a team a good while to build it. But that doesn't change the core principle.
In our case we weren't discussing how we built AIs to become MtG gods, but how some of the core principals of building one would work. Very different things. My issue was that you didn't understand what it was you were reading and yet you kept pushing on. It's probably my fault for not explaining well enough.
@TLK I generally answer properly to every question asked. There's a difference between questions and ignorance though. When someones responses are so far off the subject they either really don't get it or are trolling. I couldn't imagine (agsin, my fault) someone to not understand it so I opted for door number 2 and cut the conversation short.
To expand on that slightly more...
There's a few principals at work here. The first is the theory of how the algorithms work, a lot of people don't really care about that and it leads to really bad posts when you have to preface some basic results with pages of computer science theory just to explain what is going on. This is a mistake I've made before, so I tend to just ignore it and jump to the results now, or at the most provide a very high level overview. Generally, if people are curious or if they want to argue the details of it that's better served in PM's.
All we were really getting at here was rule based evaluations. There's nothing wrong with attempting to quantify a cards value and playing the more valuable cards. If however, you do want to quantify things there are right and wrong ways to go about it and the data entry for it is pretty extensive when done right, it's exponential actually: 100 cards is 100*100 or 10,000 pieces of data, while doubling your card pool to 200 cards is 200*200 or 40,000 pieces of data, 4x as much for a small increase to the card pool so it becomes time prohibitive to quantify anything more than a handful of cards against each other if you do it manually.
That's where the side discussion went between Echelon and I, because we've bounced ideas off of each other for the better part of a year now to come up first with a way to automate the process and then ways to better automate it. My biggest hurdle to the project these days is free time to sit down and try to do it.
There's basically three main components:
1. Figuring out how to teach the computer what the card does.
2. Creating a rule based evaluation from card to card where a comparison must pass a series of tests to rate favorably.
3. Implementing the above two.
That first point is what Echelon and I were talking about a few posts ago. I'm currently on attempt #3 for it.
My first attempt (which is what I use in my MTG simulator) doesn't work at scale as each card has to be individually programmed so it had to be scrapped for a project that's meant to include all 20,000ish Magic cards.
My second attempt was able to scale, but would have required major modifications any time Oracle changed such as new keywords or the recent damage change. So that wasn't sustainable.
My third attempt addresses that, but it's not yet written... still working on the theory. Explaining interactions to a computer like Aether Vial vs Force of Will or GSZ for Eternal Witness for something is really difficult.
The second point is what sparked the whole debate. Evaluation can work when you consider something like A vs B. For example, Thoughtseize beats Liliana of the Veil, or Lightning Bolt beats Glistener Elf. Where evaluation does not work is when you rank effects like discard = 1 point, card advantage = 1 point/card. The reason this doesn't work, is because point values don't remain static when paired with other effects. To give a small example:
Mind Rot is worth 2 points under a system that awards 1 point for each discard.
Divination is also worth 2 points under the idea that each draw is worth 1.
Now, a hypothetical 3 mana sorcery card that makes the opponent discard 1 and you draw 1 is also worth 2 points.
However, none of these 3 cards are equal to each other in power.
Lets also take another example. I won't walk through it, instead I'm just going to name cards. Rakdos's Return and Mind Twist. These are very similar cards but the power level on them is quite a bit different even though any point evaluations would result in them being fairly close.
Yes, your method has to be demonstrably better than or similar to the best humans, otherwise it doesn't provide valuable input for strategy/tactics discussions. This is the point I have been trying to make all alongPlenty have already done this. DotP did this. There is a big difference between playing the game, which is quite trivial to implement, playing the game well, and playing the game perfectly (which is what the Go and Poker AI's in that past couple years were built to do).
Why would you use a Monte Carlo method to approximate this instead of just using a Hypergeometric distribution to give you the correct answer?This example tracked the number of lands drawn at 24, 25, and 26 land
(unless your whole point was to illustrate that stochastic processes with a large sample size will always have a small inherent margin of error, in which case, fine)
Like I say, you can calculate the exact percentages for all the entries in that table, but it still wouldn't tell me whether 24/25/26 lands are correct, because all it tells me is how many lands I need to play in order to see X number of lands with Y probability in the first Z cardsPeople always go back and forth on land counts, they cut a land because they feel like they're flooding, or add a land because they were mana starved but the reality is that those types of situations play out over a handful of games, at the most 100 games. When 100,000 games doesn't even give you a solid base to conclude that 24, 25, or 26 lands are correct, and considering that land drops are the most consistent action in almost all decks, then every conclusion you make about how a deck functions is also on equally shaky ground.
This information is not totally worthless (hence why Frank Karsten's articles are so popular), but it doesn't actually tell me how many lands I should play in order to maximise my win percentage. To put it another way, it doesn't tell me what are the values of X or Y or Z that I should be aiming for.
How many lands should I play considering my mana curve? How many lands should I play considering the number of cantrips in my deck? Should I play more/fewer lands if I think my deck gets punished more for getting screwed/flooded? Can I play fewer lands if I have basic lands that don't get destroyed by wasteland? If I have my own wastelands, do I count them as lands, or as spells? If the meta has a lot of waste/stifle decks in it, do I need to play more lands? I can give contextual answers to all of these questions. Yes, my answers are based on 'shaky' human experience, but my point is that we don't currently have a computational process that accounts for these factors, so human experience is the best method we have.
You are implying that you fed a database of every modern legal card into your solver, and ab initio it spat out a list similar to what Kelvin Chew designed . This claim defies credulity.I've been testing the system for years now. The first variant played Modern Knightfall before that deck had any tournament success, and therefore no lists to base a deck skeleton off of. I was able to identify one, and that skeleton ended up getting extremely close to the deck in it's modern day form. After that I used Burn, and was able to identify cards that weren't being played that had potential. I've then added those cards and played it at tournaments, to decent success.
If what you're suggesting is instead "I came up with a deck skeleton myself, and then a computer program suggested a few tweaks based on some guidelines I decided upon myself, and in the end I got a list that was a bit worse than what a pro player came up with" then this is entirely believable, but not useful or noteworthy at all (insofar as being an endorsement of your computational system).
From someone who suggested that 100,000 games isn't enough to get good data only a few sentences earlier, it's amusing to see "I changed a couple of cards in my burn deck and won a few games with it, so the system works"
I'm not insisting that it has to be perfect, only that it has to improve upon what we can achieve without it.It's not perfect, but no one said it was. Echelon and I were just explaining to the person with their matrix that if you want to follow that approach, there's a much better way to do it which is with rule based evaluation card by card. Not a point based system for effects.
I find it bizarre that you would insist one wonky numerical method is "better" than another when you don't have any way to validate that assertion.
Simply, it seems your advice was 'Rather than try to think about what your cards do in isolation, it may be better to think about how they match up against specific cards that you expect to face'. I can't really disagree with this statement.
But it's a huge (currently: impossible) leap to go from that idea to having a system that objectively tells you the best cards to play, so pretending that such a system currently exists is misleading (or at least unhelpful).
"Proven" is a very strong wordEdit: And all of this has an actual impact on discussing deck building choices. To give an example, something you'll see very often in Nic Fit is people playing a 61st card. This tends to go against almost all conventional wisdom for deck building. However, through various methods (logic/reasoning, a simulation such as my above example, a little bit of practice, and so on) we have over the years come to the conclusion that the toolbox value of having one additional card in the deck so that it's there when you need it, outweighs the inconsistency added to the deck. I've proven this in several decks now, so it's not just a toolbox related effect. Having an out to something unexpected that you can potentially draw in your deck as an additional card is much more valuable than adhering strictly to 60 cards, and that's something I've even started applying to other decks and formats.
From MTGGoldfish results from the last 12 months it seems like there aren't any Birthing Pods, but BUG looks pretty evenly split with Junk and GB. It seems fair to tell people upfront whether your claims are based on theorycrafting or metagame data or your own testing or whatever else.For example, Pod is one of the weakest archetypes and BUG is the weakest color combination. Given the cost of the format, it's only fair to tell people up front if their build is or isn't viable.
I work in a medical research lab; it isn't my own specific area of expertise, but my colleagues do the exact kind of deep-learning for image analysis mentioned a few posts back and we frequently discuss it. What exactly is it that you think I don't understand? I haven't commented at all on the techniques you might use to invent a better system. My only problem is the claim that there is a computational approach that builds and plays decks today and that this current system can build decks at a level comparable to professional/skilled players. These claims are unverified (if not outright false) so it's wrong to suggest that the best cards/decks are found by following any sort of numerical scheme.To pile onto Brael: in software development the scale of work generally isn't considered when talking about stuff like the core workings of an AI. Consider the AI that beat pro dota players last year. This sums up how it works: It's a neural network that loses points for taking damage and gains points for inflicting damage. Sounds simple, right? Also sounds like easy to make, right? It isn't though, it took a team a good while to build it. But that doesn't change the core principle.
In our case we weren't discussing how we built AIs to become MtG gods, but how some of the core principals of building one would work. Very different things. My issue was that you didn't understand what it was you were reading and yet you kept pushing on. It's probably my fault for not explaining well enough.
Even Brael's proposed solution is incredibly limited because as-described it doesn't actually involve any machine learning. As long as the "rule based evaluation from card to card" is given to the system from its designer, and not generated by the system itself, there will always be a significant cap on the insight that can be obtained beyond what the designer already knows or believes (assuming those beliefs are even correct in the first place).
?Thoughtseize beats Liliana of the Veil
And there's a circle. Repeat ad infinitum.
The best humans aren't in this thread, and don't play this deck.Yes, your method has to be demonstrably better than or similar to the best humans
Actually, I took a bit more abstract approach for that deck initially. I started with a bunch of generic cards like combo pieces, knights, big and small 2 drops, mana dorks, rainbow lands, and so on. Then I measured how long the deck took to goldfish some number of games (it wasn't capable of playing opponents at that time) using every possible combination of cards where either 0, 2, or 4 copies of the card were played. Then I took the promising results from this (I don't remember how many in total) and tried again at more games playing 0, 1, 2 or 1, 2, 3, or 2, 3, 4 copies of each card depending on where the results suggested the optimal point was. After that, I started filling the generic placeholder cards in with real cards that fit the necessary criteria. The end result wound up being extremely close to what was eventually developed as the stock build for the deck.You are implying that you fed a database of every modern legal card into your solver, and ab initio it spat out a list similar to what Kelvin Chew designed
In the absence of professional opinion, which really isn't worth much in the first place since pro's routinely disagree with each other... then you have to use something that can better inform your choices. Also, I never claimed it played at the level of a professional, in fact I outright seek to avoid that. There's systems for that too, such as using a Monte Carlo method to always make the best play, but those systems are too computationally complex and leave the data prone to the variance in the game. Instead I use a simple heuristic and rate against that. This will not result in perfect play but it does quickly create a large volume of good plays, which can then be analyzed for statistical trends in order to see what is and isn't working... for example, answering the question of optimally sequencing Goblin Guide vs Monastery Swiftspear.
And, I think you're confused, maybe I didn't explain it enough. My AI actually does use machine learning, it encompasses genetic algorithms to include cards from the card pool and to weight the rules to play a card (all guided by a heuristic of evaluating turns until someone wins). As I said though, this system isn't scaleable. My system plays games very quickly, approximately 10 per second on the computer I use to run it. But, using genetics introduces an exponential increase in the necessary runtime to have sufficient data. As a result, my AI which had to be programmed card by card, then card pools implemented deck by deck, then build the decks from that pool, and then get a sufficient amount of data that you can harvest exactly what you want takes longer when using them... thus a better system is needed.
The system taught me quite a bit about deck building, specifically land counts and card diversification, and it's a fun project but it doesn't solve the initial problem which is getting cards into the card pool in the first place (the eventual goal is to hook up 8 of them, give them any arbitrary draft format, and iterate on the drafting and playing process until any such format can be quickly solved by doing nothing more than providing a list of cards in that format).
That's where this rule based evaluation comes in which is not AI but rather automated data entry. That data entry then feeds back into steps 1 and 2 for the AI which are finding what a card does (and circumventing the need to program them individually), and building limited card pools. From there it can go back to automated game play in order to playtest suggestions.
Tournament report for yesterday, 6/26/2018 at the Summer 1K at The Complex, Scarborough, Maine. I ended up going 3-3, beating Aggro Loam, TES, and Turbo Depths. I lost to UW Stoneblade, Esper Stoneblade, and Elves.
Round 1 - UW Stoneblade (Loss, 0-2)
Game 1 is grindy, I misalign him to Miracles, lose a grindy game where he ultimates me with Jace and I can't find an answer. I drew zero cabal therapies, but I slowed him down significantly with Doran with his TNN. Got many laughs, and some cred for being ballsy. G2 I lose to myself; he has a Leyline of sanctity, TNN, and Jace on the table. I draw To the Slaughter, get a boner, and go for it. Because he is an illegal target I have to sacrifice my own Sigarda. From there I just can't recover, even though I get close. Lose to Jace ultimate again. Tons of interactions with Nic Fit that I am learning on the fly!
Round 2 - TES (Win, 2-0)
Game 1 he has a super slow start and I steal it from him. He actually didn't board out his Dark Petition, we call a judge, we can't really resolve it (no longer a game loss), he ends up being 2 goblins short and I get him down with Rhinos. G2 I chain Duress, Therapy, Canonist and win easily with a top-decked Doran in 2 turns. He rolls his eyes at Doran, I laugh inside with glee. Suck it, boi.
Round 3 - Elves - (Loss, 1-2)
I lose g1 to a quick combo, dying with roughly 33 power on the table. G2 I therapy the shit out of his hand, he still gets a Hoof, beats me to 8. I make a hard call and play Painful Truths for 3, rip a Path, he's empty, I draw into Rhinos with Sylvan Library. G3 is much like g1 but I end up landing a Canonist, which hurts me more than him because I can't do Therapy+flashback same turn. It slows him down about 2 turns, but it isn't enough for me. I get a Deed down, but he's an old pro at Elves and doesn't overextend. I die to a million after a Hoof shows up.
Round 4 - Aggro Loam (Win, 2-0)
He starts on Chalice, which blanks a few cards, but I draw hot with Abrupt Decay for his Library, I deed away Mox, 2 Chalices, and a Confidant. I finish up with a Rhino and Scooze, which is at 5/5. G2 goes much the same, I'm in control the whole time and Rhino just ends it quick. Didn't even have to GSZ for Pridemage.
Round 5 - Esper Stoneblade (Loss, 1-2)
Game 1 is a grindfest, I deed away his board, I'm stuck with just a Pridemage while he has a TNN and 4 Souls tokens. I rip a Rogue's Passage off the top, he's at 2, I attack for 3 unblockable. His face was priceless! Cheering was heard from the common folk. G2 he outgrinds me, once again due to Jace feeding him good stuff. I have a ton of answers, including 2 deeds in hand. He pithing needle's my Deed, I gsz for my Pridemage...which I somehow sideboarded out! What the fuck is wrong with me? It must have been a mistake because I know short of being white-girl wasted I would never board out my Pridemage. I lose to TNN with a BSKull equipped. G3 is another grindfest, I get an edge but he is using Academy Ruins and Engineered Explosives on my stuff. I Vindicate the Ruins but it's too late; he gets there about 2 minutes before the round ends. Super fun games, he's a good friend of mine at the shop. The pridemage error cost me the match.
Round 6 - Turbo Depths (Win, 2-1)
G1 he's slow, I deed away Safekeeper and Pithing Needle, rhino his ass for a ton. G2 he gets a fast combo, t4 I think, and I'm dead even though I have Sigarda to block. He plays Sejiri Steppe and I'm dead. G3 he has another slow start, but he has the Hexmage in hand for a win, I therapy it away, land a Rhino to put me at 21 life. He stalls for a turn or two while I beat his face in with Rhino and drain him with Deathrite. He's at 2 life, he combos, attacks me to one, but I have a Deathrite Shaman to finish off. Winning at 1 life and being happy about it.
Best cards: Liliana, the Last Hope, Siege Rhino, Painful Truths, Sylvan Library.
Worst cards: Pithing Needle was fairly lackluster all day, I only boarded it in against Depths combo. Deed + Needle is awkward when both are good in a matchup but you kill your own Needles. Rethinking that plan, maybe it's time to figure out a different option.
Cards I wish I had: golgari charm in the sideboard. If I could have chained Charm into Deed I think Elves would be a lot more winnable. The value against the Esper Stoneblade deck would have been insane as well, and it handily dinks off my Explorers for mana. I also wish I had a Reclamation Sage to bring in from the sideboard. I may drop the Needles for those two cards in the future.
So overall, not a bad outing, just a mediocre one while I learn the ins and outs of the deck (which is super sweet, btw.) I also traded into a Japanese Phyrexian Tower and other great stuff. I'll be sticking with this deck for a while, it's definitely super fun, and surprisingly good against the fair decks. Some better sideboard options and more practice should help me tune it up. Doran did what I wanted, will likely stick with him. Suggestions for how to do different lines (I can provide more matchup details) would be awesome, but overall I think there's just so much the deck can do it takes a while to learn. It isn't a deck I can jump onto and have immediate success, that's for sure, but's its fun and good enough to put in the time.
Thanks for reading!
Brainstorm Realist
I close my eyes and sink within myself, relive the gift of precious memories, in need of a fix called innocence. - Chuck Shuldiner
I keep on hearing that there's a nic fit list that focuses on planeswalkers with academy rector. Does anybody have a good example of this list?
Arena rector.
Brainstorm Realist
I close my eyes and sink within myself, relive the gift of precious memories, in need of a fix called innocence. - Chuck Shuldiner
There are currently 1 users browsing this thread. (0 members and 1 guests)
Reply With Quote