Did a bit more testing. Forgotten Cave was fantastic in games that went to turn 6+, and less so before that. Burns main plan is to win on 4 though so something that doesn't pay off until 6 isn't where you want to be. In general it's power level was above Gitaxian Probe and Preordain but below Serum Visions (yes, Preordain is weaker, atleast in this program).
So with that in mind, it's probably something to keep in mind for a random slower deck at some point in the future. It's too bad they aren't Modern legal.
From there I went back to testing fetchlands as I had previously only done that as a goldfish rather than with an opponent. Like the previous test 12 fetchlands equated to wanting +1 land in the deck. So 21 Mountains 2 Barbarian Ring is optimal, but with fetches it's actually 11 Fetch, 11 Mountain, 2 Barbarian Ring.
Worth pointing out. While the fetch deck was faster by nearly half a turn it's overall win rate is lower. Basically, this means it was winning the T3/T4 games more often but losing the T5+ games more often. So when tuning your deck, Mountains are stronger in a slow meta but fetches are stronger in a fast one.
I've been idly reading this forum for a bit as a burn player. Recently though I've been playing a different kind of modern burn deck, Esper Mill. The current lists are in a wide array of variety and no list is super set in stone. It'd be interesting to see what the actual effectiveness of each card is, and what the brute force numbers are. I.E. does mesmeric orb do more than mind funeral or glimpse the unthinkable? How often do we have archive trap when the opponent is fetching? How often is our Jace's Phantasm buffed on turn 2 or 3? How many cards does the average Hedron Crab get? What turn do we reliably cast Visions of Beyond as Ancestral Recall?
Some of the logic seems to be the same, and most of your code may be reusable if you'd like to take on a variation of your project. I dunno, maybe you're sick of looking at red cards
This is my list from a few weeks ago, if you're interested: Esper Mill
I've been idly reading this forum for a bit as a burn player. Recently though I've been playing a different kind of modern burn deck, Esper Mill. The current lists are in a wide array of variety and no list is super set in stone. It'd be interesting to see what the actual effectiveness of each card is, and what the brute force numbers are. I.E. does mesmeric orb do more than mind funeral or glimpse the unthinkable? How often do we have archive trap when the opponent is fetching? How often is our Jace's Phantasm buffed on turn 2 or 3? How many cards does the average Hedron Crab get? What turn do we reliably cast Visions of Beyond as Ancestral Recall?
Some of the logic seems to be the same, and most of your code may be reusable if you'd like to take on a variation of your project. I dunno, maybe you're sick of looking at red cards
This is my list from a few weeks ago, if you're interested: Esper Mill
Outside of a few small changes the project is on hold until December, I won't have free time to work on it until then but I do have a few ideas. Depending on my free time I might make a few tweaks before that point (I would like to make the opposing AI a bit stronger, and shift some internal components of the program around).
The Esper Mill question is interesting and it's a similar concept to answer (I'll add it to the list of decks I'm going to eventually build) but I'm staying away from Modern for now. In Legacy I can pretty easily just fetch rainbow basics because there's no shock damage on the mana. In order to implement Modern I have to build a system to differentiate between shock/tapped shock/basic fetches, and that requires evaluating color tapping. I've got an algorithm to (mostly) do so written down in a notebook and it's something I want to work on over the winter but until I do that I can't really implement Modern decks because shock damage is super relevant to the point that it's the most defining aspect of the format.
I'm very interested in this project and these results and would be glad to help in some way, either with coding or structuring the analysis to answer meaningful questions.
I've read through a few of your posts, but will take more time to read through all the analysis later.
What I find most interesting about this research are the results that show very little difference between 22 and 27 lands in many simulations, I guess most of those were not attributed to known bugs? I wonder if that is purely burn related, since it's a very simple deck to model and it can mostly be played algorithmically, or should that carry over to more complex decks too? Critical mass type combo decks not likely, but what about control, or control-combo? Even more cycling lands or more realistically, Brainstorm/Ponder effects or Sylvan Library kind of do the same. Make you not miss a land drops, but can be cashed in for value cards when needed.
Sorry, I'm not very active in this thread right now. I don't seem to get alerts when people post to it and with my classes I don't have much time to alter it beyond adding new cards, which for the most part is pretty simple. There's a couple things I want to do over winter break with it though.
The land count issue is twofold. The first is that more lands really just is more consistent. The other issue has to do with the number of cards drawn. You need high land counts in order to curve out. If you can dump cards into card draw though you need fewer. I think the exact count is higher than what most people play in their low land+high cantrip decks though. The basics are in place for my Delver deck, that's something I'm going to dig into more in about a month and a half.
I'm very interested in this project and these results and would be glad to help in some way, either with coding or structuring the analysis to answer meaningful questions.
I've read through a few of your posts, but will take more time to read through all the analysis later.
I'm trying to do all the coding myself or possibly with a friend or two IRL who are interested, I'm an ok coder but I still have a lot to learn, and it's definitely messy so I want to get all the practice in that I can. I'm very open to being provided questions and if necessary the structure to answer them though.
As I've said before, there's very little development going on with this while I'm busy with the semester but some changes will happen over my winter break. Recently though I decided to put the program to the test, and I entered a 12 week long weekly Legacy league in paper. The deck I'm playing was built entirely by my program, including some cards I added in for it to predict the results like Wild Nacatl (it didn't like it) and Scab-Clan Berserker, which it suggested I play 2 of. So far the generated deck has been pretty good (I built the sideboard though, not the computer).
Private Mod Note
():
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
So with that in mind, it's probably something to keep in mind for a random slower deck at some point in the future. It's too bad they aren't Modern legal.
From there I went back to testing fetchlands as I had previously only done that as a goldfish rather than with an opponent. Like the previous test 12 fetchlands equated to wanting +1 land in the deck. So 21 Mountains 2 Barbarian Ring is optimal, but with fetches it's actually 11 Fetch, 11 Mountain, 2 Barbarian Ring.
Worth pointing out. While the fetch deck was faster by nearly half a turn it's overall win rate is lower. Basically, this means it was winning the T3/T4 games more often but losing the T5+ games more often. So when tuning your deck, Mountains are stronger in a slow meta but fetches are stronger in a fast one.
Some of the logic seems to be the same, and most of your code may be reusable if you'd like to take on a variation of your project. I dunno, maybe you're sick of looking at red cards
This is my list from a few weeks ago, if you're interested:
Esper Mill
And apparently I've changed my name: Ugh
Outside of a few small changes the project is on hold until December, I won't have free time to work on it until then but I do have a few ideas. Depending on my free time I might make a few tweaks before that point (I would like to make the opposing AI a bit stronger, and shift some internal components of the program around).
The Esper Mill question is interesting and it's a similar concept to answer (I'll add it to the list of decks I'm going to eventually build) but I'm staying away from Modern for now. In Legacy I can pretty easily just fetch rainbow basics because there's no shock damage on the mana. In order to implement Modern I have to build a system to differentiate between shock/tapped shock/basic fetches, and that requires evaluating color tapping. I've got an algorithm to (mostly) do so written down in a notebook and it's something I want to work on over the winter but until I do that I can't really implement Modern decks because shock damage is super relevant to the point that it's the most defining aspect of the format.
I'm very interested in this project and these results and would be glad to help in some way, either with coding or structuring the analysis to answer meaningful questions.
I've read through a few of your posts, but will take more time to read through all the analysis later.
Sorry, I'm not very active in this thread right now. I don't seem to get alerts when people post to it and with my classes I don't have much time to alter it beyond adding new cards, which for the most part is pretty simple. There's a couple things I want to do over winter break with it though.
The land count issue is twofold. The first is that more lands really just is more consistent. The other issue has to do with the number of cards drawn. You need high land counts in order to curve out. If you can dump cards into card draw though you need fewer. I think the exact count is higher than what most people play in their low land+high cantrip decks though. The basics are in place for my Delver deck, that's something I'm going to dig into more in about a month and a half.
I'm trying to do all the coding myself or possibly with a friend or two IRL who are interested, I'm an ok coder but I still have a lot to learn, and it's definitely messy so I want to get all the practice in that I can. I'm very open to being provided questions and if necessary the structure to answer them though.
As I've said before, there's very little development going on with this while I'm busy with the semester but some changes will happen over my winter break. Recently though I decided to put the program to the test, and I entered a 12 week long weekly Legacy league in paper. The deck I'm playing was built entirely by my program, including some cards I added in for it to predict the results like Wild Nacatl (it didn't like it) and Scab-Clan Berserker, which it suggested I play 2 of. So far the generated deck has been pretty good (I built the sideboard though, not the computer).