Friday, December 18, 2015

Announcing www.knowthemeta.com

Finally, my new website www.knowthemeta.com is online!


It parses tournament, deck and ranking information from Acoo.net and Stimhack.com and generates statistics. What kind of information does it provide for Netrunner players?

  • Faction percentage over time (expansions)
  • For each data pack / expansion
    • Faction and identity percentages for top 30% and all decks
    • Most used cards per data pack / expansion and tournament cardpools
    • For each identity
      • Comparing decks in a deck similarity matrix
      • Listing cards for each deck
      • Statistics of used cards


Any ideas or feedback is welcome.

Sunday, July 26, 2015

Analyzing deck archetypes with MATH

I started building this programming / math / Netrunner project of mine. The goal is to analyze huge number of Netrunner decklists and categorize them into archetypes using mathematical statistics.

What do I have so far?
  • reading card stats from netrunnerdb.com API
  • reading tournament and decklist data from acoo.net
  • currently using ugly command line input/output, totally incomprehensible to anyone but me
  • ability plot decklist similarity using mathematical statistics
What do I mean by the last point? Let's take all the Near Earth Hub decks with the Breaker Bay legal poolset from acoo.net. After putting all the data I got into a plot graph, I get this:

31 Near Earth Hub decks

What is this? This is a 2-dimension representation of deck similarity. Each dot is a deck. The distance of these dots represents the similarity of decks. If two decks are similar, they will be close to each other. If there are lot of differences, they will be far apart.

3 main deck archetypes as clusters of dots

After examining the data, it is clear that similar decks form clusters in this graph. These clusters are actual deck archetypes. Cool, huh? :)

So what's next? I will be expanding this project, adding functionality, hopefully a usable web UI. I'm totally unfamiliar with front-end technologies, so any suggestions are welcome. The current code is quite crappy, hard-to-use, hard-to-understand, please don't ask for it. I will be sharing my results here, if I make some new achievements.

Sunday, April 27, 2014

Analyzing Andromeda

In this article I'm analyzing the card pool of tournament winning Andromeda runner decks. To do this, I'm using the deck lists provided by Stimhack.

I'm limiting the decks to the "Double Time" and "Fear and Loathing" time period. It still gives a relatively big amount of data, and the cards of the last data pack didn't really make it into the most popular builds. (If you ask me Andromeda builds haven't really changed much since "Opening Moves".)

Andromeda decks are quite uniform, usually they don't show as much variance as other identities, so they are ideal for statistical analysis. There is only one build that differs  from the standard Andromeda deck, that is the Blackguard build. I will filter these out, and it still leaves us with 44 decks out of the 47 in total.

I'm providing two data after each card in brackets. Average number and percentage of being used in decks. For all the cards, see the summary table at the end of this article. So, let's see the most popular cards broken down by type!

Events

This includes burst economy and awesome criminal events.
We have Account Siphon [2.93; 100%], Dirty Laundry [2.93; 100%] and Sure Gamble [2.93; 98%] as the top money making cards. Usually Account Siphon by itself convinces people to play criminal and makes corporations grind their teeth. It's one of the best cards ever, many players consider it overpowered.
Other awesome criminal events include Special Order [2.77; 100%], Emergency Shutdown [2.36; 98%], Inside Job [2.22; 95%], Hostage [1.09; 57%] and Easy Mark [0.98; 39%]. Tutoring for the right breaker or contact helps criminals a lot.

Forged Activation Order [0.62; 32%] and Infiltration [0.53; 30%] do not see as much play as they used to.

Hardware

Desperado [2.80; 100%] is a must have and sadly, you better have three of it. In my experience, this is the card that nets you the most credits. I often mulligan if I cannot see it in my opening hand. Plascrete Carapace [2.36; 95%] is also an auto-include because of Scorched Earths and Punitive Counterstrikes. Usually two is enough because of Andromeda's ability. R&D Interface [1.84; 84%] is needed to maintain a lock on R&D, which is quite important when playing against a fast advance corporation. Other hardware are rarely ever used.

Resources

Andromeda decks usually have a couple of Connection type resources. Kati Jones [1.66; 89%] provides the runner with credits. As for additional card draw, players tend to use one of the following: John Masanori [0.34; 30%], Professional Contacts [0.34; 34%] or Mr. Li [0.27; 20%].
Same Old Thing [1.55; 64%] adds some event recursion (Account Siphon, Emergency Shutdown, Inside Job, Special Order) and Daily Cards [1.09; 41%] can be used for getting ever more credits.

ICE Breakers

Since sentries can be dangerous these days, Faerie [2.07; 84%] sees quite some play. Other cheap breakers come from the anarch suite: Corroder [1.80, 100%], Yog.0 [1.07, 86%] and Mimic [0.95; 84%]. Please note that these breakers depend highly on Datasuckers.
Femme Fatale [0.77; 70%] is there to handle troublesome ICE (i.e. Tollbooth) and can also break sentries if needed (do not forget!). Crypsis [0.75; 61%] is usually a quick access through any ice (even Chimera), but consumes a lot of credits and clicks. These two programs might help you out if other breakers have been trashed or discarded due to damage. Power Shutdown can be a real pain to the criminals, since they have no program recursion on their own (Clone Chip, Test Run, Scavange).

Other Programs


The only program worth mentioning here is Datasucker (2.61; 89%). It's just essential if you are running Yog.0 or Mimic, and it also cuts down the cost of breaking ICE. It works well with the expensive-to-strenghten breakers: Femme Fatale and Crypsis. Any other programs are marginal.

Data on all cards used

titlesetusedAIU 1
Events
Account SiphonCore100%2.93
Dirty LaundryCreation and Control100%2.93
Special OrderCore100%2.77
Sure GambleCore98%2.99
Emergency ShutdownCyber Exodus98%2.39
Inside JobCore95%2.39
HostageOpening Moves57%1.95
Easy MarkCore39%2.44
Forged Activation OrdersCore32%2.00
InfiltrationCore30%1.67
Lawyer Up True Colors16%2.13
BlackmailFear and Loathing11%1.64
Quality TimeHumanity's Shadow11%1.82
Levy AR Lab AccessCreation and Control7%1.00
StimhackCore7%1.00
Déjà VuCore2%2.00
Satellite UplinkTrace Amount2%2.00
Test RunCyber Exodus2%2.00
The Maker's EyeCore2%2.00
DieselCore2%1.00
Hardware
DesperadoCore100%2.80
Plascrete CarapaceWhat Lies Ahead95%2.48
R&D InterfaceFuture Proof84%2.19
HQ InterfaceHumanity's Shadow9%1.22
Clone ChipCreation and Control7%1.57
Dyson Mem ChipTrace Amount7%1.29
e3 Feedback ImplantsTrace Amount5%1.50
Lemuria CodecrackerCore2%3.00
Akamatsu Mem ChipCore2%2.00
Silencer 2Double Time2%2.00
DoppelgängerA Study in Static2%1.00
Resources
Kati JonesHumanity's Shadow89%1.87
Same Old ThingCreation and Control64%2.42
Daily CastsCreation and Control41%2.66
Professional ContactsCreation and Control34%1.00
John MasanoriOpening Moves30%1.13
Mr. LiFuture Proof20%1.35
Armitage CodebustingCore11%2.09
DecoyCore7%1.57
Compromised EmployeeTrace Amount5%3.00
Bank JobCore5%1.50
New Angeles City HallFuture Proof2%2.00
Access to GlobalsecCore2%1.00
Raymond FlintMala Tempora2%1.00
ScrubberA Study in Static2%1.00
The Helpful AIWhat Lies Ahead2%1.00
The SourceCreation and Control2%1.00
Underworld ContactA Study in Static2%1.00
ICE Breakers
CorroderCore100%1.80
Yog.0Core86%1.24
FaerieFuture Proof84%2.46
MimicCore84%1.13
Femme FataleCore70%1.10
CrypsisCore61%1.23
NinjaCore20%1.00
KnightMala Tempora7%2.00
Gordian BladeCore7%1.00
ZU.13 Key MasterWhat Lies Ahead5%1.50
Deus XA Study in Static2%2.00
TorchMala Tempora2%1.00
Other Programs
DatasuckerCore89%2.93
Sneakdoor BetaCore9%1.78
CrescentusA Study in Static5%1.50
Savoir-faire 2Double Time2%3.00
Magnum OpusCore2%2.00
ParasiteCore2%2.00
DjinnCore2%1.00
KeyholeTrue Colors2%1.00
Self-modifying CodeCreation and Control2%1.00
1 - average if used
2 - card of Double Time, actual usage could be higher

Update: I just realized that mplain has created similar statistics for all store championship decks. Check it out!



Thursday, April 24, 2014

Intro

This blog is dedicated to analyzing the most successful decks and winning trends of the card game Android: Netrunner.

I will be crunching numbers from Db0's OCTGN statistics and looking up info from Stimhack's Tournament Winnning Decklists.

Some disclaimers before we begin:
  • These are statistics. Always take them with a grain of salt. I'm trying to get as big sample size as I can, but sometimes it's not possible.
  • If you just hate statistics and have a strong opinion that you know better what is good and not, please don't take it out on me.
  • Luck is a factor of the game. It can happen that a bad or mediocre deck wins because of pure luck.
  • Skill is part of the game. It's possible that a great player wins a tournament with a mediocre deck. In the OCTGN statistics, I will try to filter out inexperienced and bad players.
  • Surprise can do a lot. Sub-optimal but unexpected cards can win you the game. This can distort the statistics a bit. It also means that you can have surprisingly good results without using the "average" winning deck.
  • Creating and trying out experimental, new decks are fun. Though you might win appreciation of the opponents with these kind of decks, these are often not doing as good as the standard, tournament-winning, boring ones. Usually I will talk about the optimal, unimaginative decks, sorry. I know what you are thinking. I'm trying to convince you to sell your soul, and use the most efficient machine of cards to win. If you a infuriated by such thoughts, please leave.
OK, we have these out of the way, first article coming soon! :)