Fixed duplicates in commander generated decks

[kojotak]

Possible fix for #9985

I don't know why, but Pir, Imaginative Rascal loaded from CardRelationMatrixGenerator.cardPools has collectorNumner N.A.. That is not the case if loading from StaticData.instance().getCommonCards() (this returns 11 and 1050). The N.A. collectorNumber results in equals method returning false in an attempt to remove this card from the deck, causing Pir, Imaginative Rascal beeing duplicated (in main deck and as a commander partner).

With this change, the card generated deck for Toothy, Imaginary Friend can be loaded:

[Hanmac]

@Agetian might need to look into why it got the wrong CN?

[Agetian]

Yeah, not sure why it doesn't return the proper CN :/ A deeper look into this might be needed but not sure who would be best suited for the job?

[kojotak]

I went a little bit deeper. PaperCard without the CN is loaded from the deckgendecks/Commander.dat file, not from the res/editions/*.txt files. You can check that using breakpoint in CardThemedMatrixIO#loadMatrix inspecting the matrix hashMap. Don't know, yet, why the serialized card is without CN nor how to repair (rebuild) the dat file.

[Agetian]

Oh, hmm, I'll see if I can find an opportunity to regenerate that LDA file , along with the two other ones that desperately need updating.

[kojotak]

Shouldn't the Commander.dat file contain just card names? The problem will arise again when some card will have different name/edition/collector number/flags/foil/artIndex (according to equals method). The PaperCard instance (stored in CardRelationMateixGenerator.cardPools ) is used only for DeckFormat.isLegalCard() check, but this can be delayed after loading the whole card from StaticData . The generateRandomCommanderDeck() method would be a little bit slower in exchange for smaller memory footprint of cardPools (and dat files).

[Agetian]

It's a LDA model for building Commander decks based on training data, it should be regularly updated, ideally, but I'm not sure how and who did it before (I do know how the other LDA models were generated and I update those from time to time). The card information should be a separate thing, yes, and I don't understand why CN is inferred specifically from Commander.dat and not just from the card DB :/