def computeRewards(states, reward_halflife=2.0):
# reward_halflife is measured in seconds
fps = 60.0
discount = 0.5**(1.0 / (fps * reward_halflife))
kills = [isDying(state.players[0]) for state in states]
deaths = [isDying(state.players[1]) for state in states]
# print(states[random.randint(0, len(states))].players[0])
kills = processDeaths(kills)
deaths = processDeaths(deaths)
# print("Deaths for current memory: ", sum(deaths))
# print("Kills for current memory: ", sum(kills))
damage_dealt = [
max(states[i + 1].players[0].percent - states[i].players[0].percent, 0)
for i in range(len(states) - 1)
]
# damage_dealt = util.zipWith(lambda prev, next: max(next.players[0].percent - prev.players[0].percent, 0), states[:-1], states[1:])
scores = util.zipWith(lambda x, y: x - y, kills[1:], deaths[1:])
final_scores = util.zipWith(lambda x, y: x + y / 100, scores, damage_dealt)
# print("Damage for current memory: ", sum(damage_dealt))
# print("Scores for current memory: ", final_scores[:1000])
# use last action taken?
lastQ = max(scoreActions(states[-1]))
discounted_rewards = util.scanr(lambda r1, r2: r1 + discount * r2, lastQ,
final_scores)[:-1]
# print("discounted_rewards for current memory: ", discounted_rewards[:])
return discounted_rewards
def computeRewards(states, reward_halflife = 2.0):
# reward_halflife is measured in seconds
fps = 60.0
discount = 0.5 ** ( 1.0 / (fps*reward_halflife) )
kills = [isDying(state.players[0]) for state in states]
deaths = [isDying(state.players[1]) for state in states]
# print(states[random.randint(0, len(states))].players[0])
kills = processDeaths(kills)
deaths = processDeaths(deaths)
# print("Deaths for current memory: ", sum(deaths))
# print("Kills for current memory: ", sum(kills))
damage_dealt = [max(states[i+1].players[0].percent - states[i].players[0].percent, 0) for i in range(len(states)-1)]
# damage_dealt = util.zipWith(lambda prev, next: max(next.players[0].percent - prev.players[0].percent, 0), states[:-1], states[1:])
scores = util.zipWith(lambda x, y: x - y, kills[1:], deaths[1:])
final_scores = util.zipWith(lambda x, y: x + y / 100, scores, damage_dealt)
# print("Damage for current memory: ", sum(damage_dealt))
# print("Scores for current memory: ", final_scores[:1000])
# use last action taken?
lastQ = max(scoreActions(states[-1]))
discounted_rewards = util.scanr(lambda r1, r2: r1 + discount * r2, lastQ, final_scores)[:-1]
# print("discounted_rewards for current memory: ", discounted_rewards[:])
return discounted_rewards
def processDeaths(deaths):
return util.zipWith(lambda prev, next: (not prev) and next,
[False] + deaths[:-1], deaths)
def processDeaths(deaths): return np.array(util.zipWith(lambda prev, next: float((not prev) and next), deaths, deaths[1:]))
def processDamages(percents): return np.array(util.zipWith(lambda prev, next: max(next-prev, 0), percents, percents[1:]))
def processDeaths(deaths): return util.zipWith(lambda prev, next: (not prev) and next, [False] + deaths[:-1] , deaths)