Blackjack-RL/BJRL.py at master · willbelcher/Blackjack-RL · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits import mplot3d

import blackjack


class BJRL():

    A = np.empty(2, dtype=int)
    gamma = 0.5
    epsilon = 1
    epsilon_decay = 0.99999
    epsilon_min = 0.01
    alpha = 0.01

    def __init__(self):
        self.A[0], self.A[1] = 0, 1
        self.policy = np.full((22, 22, 2), 0.5)
        #self.N = np.zeros((22, 22, 2))
        self.V = np.zeros((22, 22))
        self.Q = np.zeros((22, 22, 2))
        self.na = len(self.A)

        self.average = []
        self.epsilon_hist = []
        self.reward_hist = []


    def train(self, num_iters=10000):
        plt.figure(1)
        plt.title("State-Value function")
        self.ax = plt.axes(projection='3d')
        self.ax.view_init(28, -131)
        self.plot()

        plt.figure(2)
        plt.title("Epsilon vs. Time")
        plt.plot(0, self.epsilon)
        input()

        b = blackjack.blackjack()
        t = 0

        for it in range(num_iters+1):
            history = self.run_episode(b)
            self.evaluate(history)
            self.epsilon = max(self.epsilon*self.epsilon_decay, self.epsilon_min)

            if t == 0:
                t=10000

                self.plot()
                av_reward = np.sum(self.average)/t
                print("[*] Iteration:", it)
                print("[*] Average reward:", av_reward)

                self.average=[]
                self.epsilon_hist.append(self.epsilon)
                self.reward_hist.append(av_reward)

            t-=1

        self.epsilon = 0

        for i in range(len(self.Q)):
            for i2 in range(len(self.Q[0])):
                self.improve([i, i2])


    def run_episode(self, b):
        history = []

        b.deal()

        while True:
            agent_sum, dealer_sum = b.get_state()

            action = np.random.choice(self.A, p=self.policy[agent_sum, dealer_sum])

            reward, over = b.choose_action(action)

            history.append([reward, agent_sum, dealer_sum, action])
            self.average.append(reward)
            #self.N[agent_sum, dealer_sum, action] += 1

            if over:
                break

        return history


    def evaluate(self, history):

        for _ in range(len(history)):
            total, index = self.disc_value(history)

            self.Q[index[0], index[1], index[2]] = self.Q[index[0], index[1], index[2]] + self.alpha * (total - self.Q[index[0], index[1], index[2]])
            self.V[index[0], index[1]] = self.V[index[0], index[1]] + self.alpha * (total - self.V[index[0], index[1]])

            self.improve(index)


    def improve(self, index):
        best_action = np.argmax(self.Q[index[0], index[1]])
        self.policy[index[0], index[1]] = np.full(self.na, (self.epsilon/self.na))
        self.policy[index[0], index[1], best_action] = (self.epsilon/self.na) + 1 - self.epsilon

    def disc_value(self, history):

        v, index = np.split(history.pop(0), [1])
        v = float(v)

        for ind, val in enumerate(history):
            v+=(self.gamma ** ind) * val[0]


        return v, index

    def plot(self):
        plt.figure(1)
        plt.cla()
        V = self.V[2:,2:]

        (x, y) = np.meshgrid(np.arange(V.shape[0]), np.arange(V.shape[1]))
        self.ax.plot_wireframe(x, y, V, cmap='binary')

        plt.draw()
        plt.xlabel("Dealer_sum")
        plt.ylabel("Agent_sum")
        plt.pause(0.000001)

        plt.figure(2)
        plt.cla()
        plt.subplot(2, 1, 1)
        plt.plot(np.arange(len(self.epsilon_hist)), self.epsilon_hist)
        plt.ylabel("Epsilon")

        plt.subplot(2, 1 , 2)
        plt.plot(np.arange(len(self.reward_hist)), self.reward_hist)
        plt.xlabel("Iterations")
        plt.ylabel("Average Reward")
        plt.pause(0.000001)

    def save(self, filename="default"):
        np.save(filename, self.policy)
        print("\n[*] Model save successfully\n")