Collectible Card Game Headquarters

The page generated is quite huge but empty after the summary table.

I did download the csv file but not sure if there should be the complete info (aka the rarity drawing, etc...), I only have a stat table. I will upload it asap.

Edit: Uploaded the report

Edit2: SDA online does not work in Firefox 38.05, even with uBlock and Ghostery disabled, it work fine with Chrome though.
I don't either get the full stat for bulk upload online

Solo upload I get stats:


Bulk I get no full stats:

Ah, ok... I had misunderstood what you meant.

In fact, what you describe is expected behaviour. The plots (and the table with the full list of cards) do not get generated in "bulk analysis" mode: just the summary statistics and the "comparison with the reference". If you want, I can add an option to generate them as well (it's just that... if you analyse 55 decks, you will get 165 graphs... if that is not a problem for you, then it's not a problem for me either). In fact, I've though now of a way of adding that option in a more elegant way... I'll try to implement that when I have a bit of time...

I also intended to add some sort of "power level estimate" for each deck, in the summary table, but I'm still working on that problem.

Regarding Firefox not working... that's strange, since I am running Firefox 38.0.5 and it works on my side (as long as Javascript is activated, of course), both for the online and offline version. I don't use Ghostery or uBlock, but NoScript and RequestPolicy don't seem to interfere (again, as long as you whitelist the site). When you say it doesn't work... how does it manifest itself? Does it work in the offline version?

Ah sorry, forgot to add what work wrong with Firefox : hitting the button to select a deck file from computer do nothing, I have also NoScript disabled so don't know what's wrong

I also intended to add some sort of "power level estimate" for each deck, in the summary table, but I'm still working on that problem.

Well, in fact that is the feature that I hoped to be integrated, some sort of power level according to deck ruling, rarity and AI base value, would be very nice to see it in.

I concede that mass bulking would make full stats difficult to read, so what about a power level comparison summary table? With that you could appropriately compare enemy decks and decide which deck is too much/not enough powerful for which enemy...

stassy wrote:Ah sorry, forgot to add what work wrong with Firefox : hitting the button to select a deck file from computer do nothing, I have also NoScript disabled so don't know what's wrong

That's strange :\ do you get the same in the offline version? If yes, it could be due to the fact that the online version does cross-site scripting, while the offline version doesn't.

stassy wrote:Well, in fact that is the feature that I hoped to be integrated, some sort of power level according to deck ruling, rarity and AI base value, would be very nice to see it in.

I concede that mass bulking would make full stats difficult to read, so what about a power level comparison summary table? With that you could appropriately compare enemy decks and decide which deck is too much/not enough powerful for which enemy...

Yes, I really should try to get that done. A "power level" based on the AI base value is trivial to calculate... the only problem is that the valuation is very draft-oriented (e.g. tribal cards and other cards that don't work very well in Limited are heavily penalized): if you want, I can add that quickly (since it's already implemented anyway)... I guess having a not-very-good measure is better than having no measure? A "power level" based on card rarity seems also easy to implement (though I'm not 100% sure of how powerful "rarity" is, by itself, as a predictor of "deck power"), as long as I can find a reasonable scoring function (i.e. some way to map "rarity" to "card value").

My current idea is to model the real life price of cards as a function of "card rarity" and "AI base value", to have an idea of the effect of "card rarity" on card prices. Then, if that effect is removed (i.e. if we predict what the cost of each card would be if they were common), you probably get a decent measure of "card value" (i.e. whatever it is that makes cards cost more or less money, after accounting for "card rarity"). I just have to find some time to parse this huge dump of historical MTG card pricing data (~2GB uncompressed ) and do some quick analysis, and I might have that working... j'espère

Of course, all of this is just an approximation: it is obvious that the value of a deck is not a simple average of the single card values... there is synergy between cards which makes the relation between "card value" and "deck value" highly nonlinear (silly example: Instill Energy and Colossus of Sardia, by themselves, are crap; together, they are still crap, but much better than a simple sum of their values ).

It work offline, so maybe that's only my Firefox version which is bugged...

For the card price, what about the price data from Shandalar itself (the vendors)? I think it's automatically set on new cards so it can be also used as a value factor...except mythic rare of course :p

stassy wrote:For the card price, what about the price data from Shandalar itself (the vendors)? I think it's automatically set on new cards so it can be also used as a value factor...except mythic rare of course :p

It is automatically set, yes. In fact, Korath added a new function to Shandalar that outputs a list of card prices for all cards according to Shandalar's internal pricing function. I already have that information with me, so it would not be difficult to add to the app. The thing is... from what I've been able to determine, the AI Base Value is a much better predictor of "card value" than Shandalar's pricing function (example: if you take two cards which are the same, but one costs 1 mana and the other costs 2 mana, Shandalar's pricing function will say that the most expensive card is "better", which is obviously false).

I am currently looking at the other (real life MTG card pricing) dataset I mentioned. I will compare it will "AI Base Value" and the Shandalar pricing, to see if there is any obvious trend...

After looking at those real life card prices and so on, I think I managed to come up with something vaguely reasonable...

This deck scoring thing is active in the online version. If there aren't any big bugs, I'll release an update for the offline version as well...

The logic of the scoring system is based on stuff like "real card prices", "prices in shandalar", "ai base value", "rarity" and other stuff. The relation between "card value" and "deck value" is not linear, and is based on a robust (i.e. resistant to outliers) ensemble approach, so don't worry if some cards seem misclassified. The scaling is logarithmic and follows these rules:

1) the 25% worst cards are below 0;
2) the 25% best cards are above 12;
3) a deck cannot have negative value, unless you are choosing really bad cards and/or hand-picking cards which are misclassified;
4) typical shandalar decks end up being in the 0-24 range (originally, at least);
5) it's hard to get a deck above 24, unless you are using cards which are really expensive in real life (moxes, black lotus, timetwister...);
6) only nonland cards count for deck scoring and some types of decks are still underestimated (tribal decks, any decks that involve strong card synergy... like life gain decks, combo decks, etc.);
7) the practical limits of the scale are from -50 to 100, but it is almost impossible to get a deck close to those values (unless you are doing silly things like submitting a deck of 60 black lotus).

As an example, here's the image you would get if you uploaded jiansonz's 55 decks from his original Shandalar deck pack:

As you can see, the deck classifications are generally between 0 and 24 and the decks get more or less ordered according to their tier, kinda...

Very nice, will look forward the offline version :3

I guess that for AI, combo/synergy value can be disregarded because unless it get the perfect hand (sometimes it does for a storm deck, but once per 50 game) or tinkered through the cards themselves, it won't play combo deck well.

Also would be nice for the sum deck chart to have the enemy name next to the deck filename so we can take a quick peek at the powerlevel (999 is Arzakon right? Doesn't seems to fare well :p )

Thanks for the feedback. I have prepared a new version (which is now active online), so I will shortly prepare the update for the offline version (probably today, later on).

I'm not so worried about combo decks as I am worried about "synergy decks". Tribal decks are a subset of these. The difference between "combo decks" is that "synergy decks" don't necessarily need any particular combination of cards to work... the AI usually plays well with those (at least if they are aggro-based). A good example is Merfolk Tribal... the curve is made so that opponent can almost always drop 1 to 2 merfolk per turn and almost half of them can be "lords", since merfolk have so many (Lord of Atlantis, Master of the Pearl Trident, Merfolk Sovereign). For every deck where the total deck value is more than the sum of its parts (i.e. cards), there is the potential for some mis-estimation...

The good news, though, is that is otherwise seems to work pretty ok (as you can see in the examples at the bottom of this post).

Regarding Arzakon, well... just check his deck out, really

Arzakon's POWER deck! | Open

.9 3 Badlands
.12 3 Bayou
.189 3 Plateau
.212 3 Savannah
.266 3 Volcanic Island
.258 3 Underground Sea
.241 3 Taiga
.254 3 Tundra
.252 3 Tropical Island
.216 3 Scrubland
.425 2 Hurr Jackal
.165 1 Mox Emerald
.166 1 Mox Jet
.167 1 Mox Pearl
.168 1 Mox Ruby
.169 1 Mox Sapphire
.870 4 Gem Bazaar
.76 2 Elvish Archers
.415 2 Erg Raiders
.106 2 Grizzly Bears
.161 2 Mesa Pegasus
.942 2 Pikemen
.211 2 Samite Healer
.213 2 Savannah Lions
.286 2 Will-o'-the-Wisp
.859 2 Zephyr Falcon
.841 2 Wall of Dust
.1 1 Ancestral Recall
.14 1 Berserk
.202 1 Regrowth
.249 1 Time Walk
.250 1 Timetwister
.282 1 Wheel of Fortune
.230 1 Sol Ring
.25 1 Braingeyser
.3 2 Animate Dead
.583 2 Blood Lust
.481 1 Circle of Protection: Artifacts
.33 1 Circle of Protection: Black
.34 1 Circle of Protection: Blue
.35 1 Circle of Protection: Green
.36 1 Circle of Protection: Red
.37 1 Circle of Protection: White
.409 2 Desert Twister
.65 2 Disintegrate
.88 2 Fog
.97 2 Giant Growth
.649 2 Giant Strength
.113 2 Howl from Beyond
.679 2 Immolation
.145 2 Lightning Bolt
.162 1 Mind Twist
.179 2 Paralyze
.191 2 Power Sink
.752 2 Pyrotechnics
.198 2 Raise Dead
.227 2 Siren's Call
.794 2 Spirit Link
.508 4 Mishra's Factory
.396 1 Aladdin's Ring
.580 1 Black Mana Battery
.584 1 Blue Mana Battery
.930 2 Brass Man
.484 1 Clockwork Avian
.38 1 Clockwork Beast
.485 1 Colossus of Sardia
.931 1 Dancing Scimitar
.376 1 Diabolic Machine
.318 4 Fellwar Stone
.662 1 Green Mana Battery
.500 1 Grapeshot Catapult
.176 1 Obsianus Golem
.525 1 Shapeshifter
.534 1 Tetravus
.243 2 The Hive
.538 1 Triskelion
.539 1 Urza's Avenger
.852 1 White Mana Battery
.765 1 Red Mana Battery

Sure, it has plenty of very expensive goodies... but then it packs things like CoP's, Mana Batteries, Grapeshot Catapult, Raise Dead, Pikemen, Mesa Pegasus, Zephyr Falcon, Samite Healer, Grizzly Bears, etc. I guess, even originally, the underlying difficulty of playing against Arzakon was never about him having a particularly strong deck, but about him having 99999 lives

Changes introduced:

1) I've added the character names, as you suggested.

2) In case it is not obvious, both in the table and in the graph, the decks are ordered according to the "overall" deck power level. Because the "robust measure" I was using was possibly not the best option, I have added the option to calculate the "overall" deck power level using other (simpler) measures: the first option is to simply take the mean card power level (i.e. the center of mass of the distribution), the other option is to take the typical card power level (i.e. the median, the point that separates the 50% best cards in the deck from the 50% worst cards in the deck). It is easier to understand the differences by trying the different options and seeing how the organization of the graphs... either way, it doesn't change the "global picture" that much.

3) I added a visual cue for the "character tier" (now each boxplot is colored according to the character's level... low level characters are red... medium level characters orange... high level characters yellow and then arzakon white... following a gradient): this should make it easier to visually assess if the global distribution of the deck power levels makes sense or not. This was what actually suggested to me that maybe using just "mean card power level" as a measure of deck power level might be a good option.

4) Probably the most useful part, I've added a plot to let the user assess if the group of uploaded decks seems balanced, in general. This is achieved by performing a regression of "overall deck power level" (which, as pointed out in item #2 above, can be somewhat set by the user) against "character tier", along with the calculation of correlation coefficients (pearson and kendall). The technical details are not very important, since i add a little "interpretation" to the graph... more important than statistical parameters is for the user to visually assess whether there is "balancedness" (i.e. whether, as the tiers go up, the deck power levels also tend to go up or not). From my tests, changing the "overall power level" measure doesn't affect this graph that much, surprisingly.

Example (for the same set of decks as before, but using the "mean" and "median" measures, respectively):

- using mean (seems the best option, so it is now the default)

graphs for using "mean card value" as a measure of deck value | Open

- using median

graphs for using "typical card value" as a measure of deck value | Open

Here is what happen when I upload the original Shandalar deck from ISO version under balanced mode:





War mage op, nerf plz

Can you post that deck here? I'm curious...

In such cases where a deck seems to be classified quite high or quite low, it is a good idea to analyse it by itself, where you get an estimate per card of how the scoring system scores it. For some cards, it might be slightly over-rated or under-rated (in which case, you can just ignore the scoring for that particular deck).

Anyway, I've added a small change by adding some sort of "power level guidelines" for each tier (grey triangles) and now, whenever you hover the mouse close to a point, it will identify it (so it is easier to locate which decks are atypical).

Here you go :

War Mage (Re, 4th)

.164 23 Mountain
.157 4 Mana Vault
.145 4 Lightning Bolt
.84 4 Fireball
.75 4 Earthquake
.229 2 Smoke
.295 4 Ball Lightning
.65 4 Disintegrate
.155 4 Mana Flare
.396 2 Aladdin's Ring
.282 1 Wheel of Fortune
.752 1 Pyrotechnics

.vNone
.490 2 Detonate
.869 1 Orcish Catapult
.vBlack
.490 2 Detonate
.869 1 Orcish Catapult
.vBlue
.199 1 Red Elemental Blast
.490 1 Detonate
.869 1 Orcish Catapult
.vGreen
.490 2 Detonate
.869 1 Orcish Catapult
.vRed
.490 2 Detonate
.869 1 Orcish Catapult
.vWhite
.238 1 Sunglasses of Urza
.490 1 Detonate
.869 1 Orcish Catapult

Thanks. Yeah... I guess it's the combination of cards which are either burn (and therefore, valued, in terms of "AI Base Value") and/or cards that are expensive in real life (Mana Vault, Wheel of Fortune, Mana Flare, Lightning Bolt, Fireball, Disintegrate) that makes the scoring algorithm give such a high value to that deck...

It kinda is a decent deck, though

The online analyzer doesn't seems to recognize fetch land :

;0151_nether_fiend
;
;User
;User E-Mail
;24/06/2015
;1
;4th Edition
;

.151 2 Lord of the Pit
.548 2 Yawgmoth Demon
.131 1 Jayemdae Tome
.239 18 Swamp
.410 1 Diamond Valley
.1331 2 Nuisance Engine
.55 4 Dark Ritual
.497 1 Gate to Phyrexia
.68 2 Drain Life
.1440 1 Bitterblossom
.10561 2 Halo Hunter
.11118 1 Elixir of Immortality
.2037 1 Minion of Tevesh Szat
.11839 2 Rune-Scarred Demon
.12386 1 Griselbrand
.7292 1 Kuro, Pitlord
.13931 2 Master of the Feast
.10925 1 Evolving Wilds
.1468 1 Terramorphic Expanse
.12623 1 Staff of Nin
.246 1 Throne of Bone
.3593 2 Cruel Edict
.1483 2 Diabolic Edict
.11274 2 Grasp of Darkness
.1299 1 Spirit of the Night
.7368 1 Seizan, Perverter of Truth
.11451 2 Go for the Throat
.11112 2 Dark Tutelage

Maybe what I meant by "fetchlands" was not entirely clear

I actually only meant the Onslaught and Zendikar fetchlands (i.e. the "good" fetchlands ), namely: Verdant Catacombs, Wooded Foothills, Marsh Flats, Scalding Tarn, Flooded Strand, Windswept Heath, Misty Rainforest, Polluted Delta, Bloodstained Mire and Arid Mesa.

Any other land that fetches another land can be used without limitation (including Krosan Verge, Terminal Moraine, Terramorphic Expanse, Evolving Wilds, Myriad Landscape, Flood Plain, Bad River, Rocky Tar Pit, Mountain Valley, Grasslands, Bant Panorama, Esper Panorama, Grixis Panorama, Jund Panorama and Naya Panorama).

In a nutshell, if a fetchland comes into play tapped and/or the land it fetches comes into play tapped, then there aren't any limits.

(Nice deck, by the way )