If you had access to a database with an opponents history with access to what cards were played on which turns, you could add up all the cards played and turn the stats into percentage numbers (20 times player played nubian egg, 30 times player played leper gnome, 20 times player played ardent , 10 times player played abusive sergeant, 20 times played 10 times, senjin shieldmasta - (110/100)*times said card was played each turn = %value of seeing said card).
This would need to be done per deck, and you would need to run priority checks on all decks until you can distinguish which deck your playing against (only 1 deck uses flame imp, and flame imp was played on turn one. You drop all other decks from memory at this point). For cards used across multiple decks, you would just create a co-efficient of the cards per deck (3 decks use flame imp, 2 have x2, 1 has x1. all total flame imp usage for a single deck would be total flame imp usage/2.5)
This could be done at the beginning of the match, so you would peak resources at the beginning, then just load the stats into a struct array.
However, I don't know of any way to access a random opponents play history.
