コード例 #1
0
def guess(length=4, can_repeat=False):

    # init possible answers
    possible_ans = []
    if can_repeat:
        happly(10, length, possible_ans.append)
    else:
        papply(10, length, possible_ans.append)

    def list2str(g):
        return ("".join(list(map(str, g))))

    # input alert
    print("=============================")
    print(" Input example:")
    print("     \"1 2\" stands for 1A2B ")
    print("     \"0 0\" stands for 0A0B ")
    print("     \"4 0\" stands for 4A0B ")
    print("=============================")

    # start guessing
    last_res = (0, 0)
    while not last_res == (4, 0):
        last_guess = choice(possible_ans)
        last_res = tuple(
            map(int,
                input(list2str(last_guess) + "   ?A ?B: ").split()))
        possible_ans = list(
            filter(lambda x: scoring(last_guess, x) == last_res, possible_ans))
        if len(possible_ans) == 0:
            print("No answers left. Did you input correctly?")
            return
コード例 #2
0
ファイル: guess.py プロジェクト: MiccWan/game-algorithms
 def eval_a_guess(g):
     ab_cnt = dict()
     for ans in possible_ans:
         guess_result = scoring(g, ans)
         ab_cnt[guess_result] = ab_cnt.get(guess_result, 0) + 1
     # return the group size with max number
     return (-ab_cnt[max(ab_cnt)])
コード例 #3
0
ファイル: autoPlay.py プロジェクト: MiccWan/game-algorithms
 def eval_a_guess(g):
     ab_cnt = dict()
     for ans in possible_ans:
         guess_result = scoring(g, ans)
         ab_cnt[guess_result] = ab_cnt.get(guess_result, 0) + 1
     if DEBUG:
         print("evaling", g, "return with", -max(ab_cnt.values()))
         # print(ab_cnt)
         # print(max(ab_cnt.values()))
     # return the group size with max number(negative)
     return (-max(ab_cnt.values()))
コード例 #4
0
    def dfs(possible_ans, depth=1):

        ab_cnt = {}
        # print("Waiting algorithm response...")
        new_guess = algorithm(possible_ans, all_ans)
        print("Generated guess:", new_guess)
        for ans in possible_ans:
            guess_result = scoring(new_guess, ans)
            ab_cnt[guess_result] = ab_cnt.get(guess_result, []) + [ans]

        for key in ab_cnt:
            if key == (length, 0):
                dfs_result[list2str(ab_cnt[key][0])] = depth
            else:
                dfs(ab_cnt[key], depth + 1)
コード例 #5
0
ファイル: autoPlay.py プロジェクト: MiccWan/game-algorithms
def autoGuess(secret, length=4, can_repeat=False, max_num=10, DEBUG=False):

    # print("Now tring to guess", secret)

    # init possible answers
    possible_ans = []
    if can_repeat:
        hpapply(max_num, length, possible_ans.append)
    else:
        papply(max_num, length, possible_ans.append)
    all_ans = possible_ans.copy()
    if DEBUG:
        print("finished init possible_ans", possible_ans)
        print(len(possible_ans))

    # return a filtered result after a guess
    def possible_ans_filtered_by_a_guess(possibilities, guess, result):
        return (list(
            filter(lambda x: scoring(guess, x) == result, possibilities)))

    # return the min number that an answer can remove from possible answer
    def eval_a_guess(g):
        ab_cnt = dict()
        for ans in possible_ans:
            guess_result = scoring(g, ans)
            ab_cnt[guess_result] = ab_cnt.get(guess_result, 0) + 1
        if DEBUG:
            print("evaling", g, "return with", -max(ab_cnt.values()))
            # print(ab_cnt)
            # print(max(ab_cnt.values()))
        # return the group size with max number(negative)
        return (-max(ab_cnt.values()))

    # generate a guess by Knuth's algorithm
    def get_choice(all_ans, possible_ans):
        best = -len(possible_ans)
        if best == -1:
            return possible_ans[0]
        should_guess = 0
        for ans in all_ans:
            ans_score = eval_a_guess(ans)
            if ans_score > best:
                if DEBUG:
                    print("if guess", ans, "the score will be", ans_score,
                          "which is better.")
                best = ans_score
                should_guess = ans
        return (should_guess)

    def list2str(g):
        return ("".join(list(map(str, g))))

    # start guessing
    last_res = (0, 0)
    is_first_guess = True
    steps = 0
    while not last_res == (length, 0):
        # decide what to guess
        if is_first_guess:
            # User may input the first guess
            last_guess = [0, 0, 0, 0]
            is_first_guess = False
        else:
            last_guess = get_choice(all_ans, possible_ans)
        steps += 1
        last_res = scoring(last_guess, secret)
        if DEBUG:
            print("Guessed", last_guess)
            print("Get result", last_res)
        # eliminate the possible answers with the last result
        possible_ans = list(
            filter(lambda x: scoring(last_guess, x) == last_res, possible_ans))
        if DEBUG:
            print("There are {} answers left.".format(len(possible_ans)))
        if len(possible_ans) == 0:
            print("No answers left. Did you input correctly?")
            return
    print("Guessing {} used {} steps.".format(secret, steps))

    # record the worst situation
    global worst_steps, hardest_secret
    if worst_steps < steps:
        worst_steps = steps
        hardest_secret = secret
        print("New record! Guessing {} used {} steps.".format(secret, steps))
    global total_steps
    total_steps += steps
コード例 #6
0
ファイル: autoPlay.py プロジェクト: MiccWan/game-algorithms
 def possible_ans_filtered_by_a_guess(possibilities, guess, result):
     return (list(
         filter(lambda x: scoring(guess, x) == result, possibilities)))
コード例 #7
0
ファイル: guess.py プロジェクト: MiccWan/game-algorithms
def guess(length=4, can_repeat=False):

    # init possible answers
    possible_ans = []
    if can_repeat:
        happly(10, length, possible_ans.append)
    else:
        papply(10, length, possible_ans.append)

    # return a filtered result after a guess
    def possible_ans_filtered_by_a_guess(possibilities, guess, result):
        return (list(
            filter(lambda x: scoring(guess, x) == result, possibilities)))

    # return the min number that an answer can remove from possible answer
    def eval_a_guess(g):
        ab_cnt = dict()
        for ans in possible_ans:
            guess_result = scoring(g, ans)
            ab_cnt[guess_result] = ab_cnt.get(guess_result, 0) + 1
        # return the group size with max number
        return (-ab_cnt[max(ab_cnt)])

    # generate a guess by Bulls and Cows algorithm
    def get_choice():
        best = -len(possible_ans)
        if best == -1:
            return possible_ans[0]
        should_guess = 0
        for ans in possible_ans:
            ans_score = eval_a_guess(ans)
            if ans_score > best:
                best = ans_score
                should_guess = ans
        return (should_guess)

    def list2str(g):
        return ("".join(list(map(str, g))))

    # input alert
    print("=============================")
    print(" Input example:")
    print("     \"1 2\" stands for 1A2B ")
    print("     \"0 0\" stands for 0A0B ")
    print("     \"4 0\" stands for 4A0B ")
    print("=============================")

    # start guessing
    last_res = (0, 0)
    is_first_guess = True
    while not last_res == (length, 0):
        if is_first_guess:
            last_guess = possible_ans[0]
            is_first_guess = False
        else:
            last_guess = get_choice()
        last_res = tuple(
            map(int,
                input(list2str(last_guess) + "   ?A ?B: ").split()))
        # print(possible_ans)
        possible_ans = list(
            filter(lambda x: scoring(last_guess, x) == last_res, possible_ans))
        if len(possible_ans) == 0:
            print("No answers left. Did you input correctly?")
            return