I would like to understand how the win rate was computed.
Task description only says "The participants of the competition are asked to predict win-rates of 200 Hearthstone decks described in the file testDecks.json..."
How exactly was the win rate computed for particular bot and deck? Was it computed based on games between all AI bots and all testing decks and averaged later?
If so, how many times each game (AI1, AI2, deck1, deck2) was played?
for each combination of bot/test deck, we measured the percentage of wins in games against the decks/bots from the training set. Opponents were chosen at random, but the distribution of games with particular bots was not uniform - it was similar to the distribution of games in the training set.
So for every pair bot/test deck matches were played with all decks from training set but with random bot? Is it correct?
Could you provide exact information about goal of predicion? Its 11 days before final deadline and the exact solution score evaluation is unclear.
My doubts is if all pair deck/test bot play with ALL training decks but with random bot?
Without training data directly corresponding to win-rate score modeling without this knowledge is purposeless.
for each game that was used to compute the win-rates for test records, both a bot and a deck from the training set were chosen at random.