def test_on_words(words, giver, guessers, to_print=False):
all_turns, all_winloses = [], []
random.seed(90)
for word in words:
guesser = random.choice(guessers)
try:
finished_turn, winlose, clues, guesses = play_a_word(
word, giver, guesser, to_train=False)
except ValueError:
print("ERROR ", word)
if winlose:
all_turns.append(finished_turn)
all_winloses.append(winlose)
if to_print:
print("Word: {}".format(word))
print("Clues: {}".format(clues))
print("Guesses: {}".format(guesses))
print("Win Or Lose {}".format(winlose))
#print(TO_PRINT_ATTENTION)
win_rate = np.array(all_winloses).mean()
if win_rate:
avg_turns = np.array(all_turns).mean()
else:
avg_turns = 0
to_print = [win_rate, avg_turns]
print("Play with Topic Guessers, win_rate {:.2f}, avg_turn {:.2f}".format(
*to_print))
return win_rate, avg_turns
def add_experience(giver, guesser, words):
word = random.choice(words)
try:
_, winlose, clues, guesses = play_a_word(word, giver, guesser)
except ValueError:
return
rewards = get_immediate_rewards(guesser, clues, guesses)
guesser.update_memory((clues, guesses, winlose, rewards))
def add_experience(giver, guesser, words):
word = random.choice(words)
try:
_, winlose, clues, guesses = play_a_word(word, giver, guesser)
except ValueError:
return
if isinstance(giver, PGACAgent):
rewards = get_immediate_rewards(giver, clues, guesses, topics=guesser.topics)
giver.update_memory((word, clues, guesses, winlose, rewards))
def add_experience(giver, guesser, words):
word = random.choice(words)
topic = make_new_topic(word, giver.strength_dict)
guesser.set_topic(topic)
try:
_, winlose, clues, guesses = play_a_word(word, giver, guesser)
except ValueError:
return
if isinstance(giver, PGACAgent):
rewards = get_immediate_rewards(giver, clues, guesses)
giver.update_memory((word, clues, guesses, 3 * winlose, rewards))
def add_experience(giver, guesser, words):
word = random.choice(words)
try:
_, winlose, clues, guesses = play_a_word(word, giver, guesser, to_train=True)
print(winlose, clues, guesses)
except ValueError:
return
if isinstance(guesser, ActorCriticAgent):
rewards = get_immediate_rewards(guesser, clues, guesses)
guesser.update_memory((clues, guesses, winlose, rewards))
if isinstance(giver, ActorCriticAgent):
rewards = get_immediate_rewards(giver, clues, guesses)
giver.update_memory((word, clues, guesses, winlose, rewards))
def get_two_trace(target, giver):
traces = []
guesses_set = set()
for _ in range(1000):
guesser = random.choice(guessers)
_, is_win, clues, guesses = play_a_word(target,
giver,
guesser,
to_train=False)
print(is_win, clues, guesses)
guesses = tuple(guesses)
if guesses not in guesses_set:
guesses_set.add(guesses)
traces.append(clues)
if len(traces) == 2:
break
return traces