DJL/data_generator.py at main · HengruiCai/DJL · 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
from abc import ABC
import numpy as np
import pickle
import random

class DataGenerator(ABC):
    def __init__(self, mu=0, sigma=1, lb=-1., ub=1., p=20):
        self._mu = mu
        self._sigma = sigma
        self._lb = lb
        self._ub = ub
        self.p = p

    def get_context(self):
        return NotImplementedError

    def get_reward(self, context, action):
        return NotImplementedError

    def get_mean_reward(self, context, action):
        return NotImplementedError


class DataGeneratorScena1(DataGenerator):
    def __init__(self, mu=0, sigma=1, lb=-1., ub=1., p=20):
        super(DataGeneratorScena1, self).__init__(mu, sigma, lb, ub, p)

    def get_context(self):
        return np.random.uniform(self._lb, self._ub, self.p)

    def get_reward(self, context, action):
        return (action < 0.35) * (1 + context[0]) + (action >= 0.35) * (action <= 0.65) * (context[0] - context[1]) + (action > 0.65) * (1 - context[1]) + np.random.normal(self._mu, self._sigma, 1)[0]

    def get_mean_reward(self, context, action):
        return (action < 0.35) * (1 + context[0]) + (action >= 0.35) * (action <= 0.65) * (context[0] - context[1]) + (action > 0.65) * (1 - context[1])


class DataGeneratorScena2(DataGenerator):
    def __init__(self, mu=0, sigma=1, lb=-1., ub=1., p=20):
        super(DataGeneratorScena2, self).__init__(mu, sigma, lb, ub, p)

    def get_context(self):
        return np.random.uniform(self._lb, self._ub, self.p)

    def get_reward(self, context, action):
        return (action < 0.25) + (action >= 0.25) * (action < 0.5) * (np.sin(2 * np.pi * context[0])) + (action >= 0.5) * (action < 0.75) * (0.5 - 8 * (context[0] - 0.75) ** 2) + 0.5 * (action >= 0.75) + np.random.normal(self._mu, self._sigma, 1)[0]


    def get_mean_reward(self, context, action):
        return (action < 0.25) + (action >= 0.25) * (action < 0.5) * (np.sin(2 * np.pi * context[0])) + (action >= 0.5) * (action < 0.75) * (0.5 - 8 * (context[0] - 0.75) ** 2) + 0.5 * (action >= 0.75)


class DataGeneratorScena3(DataGenerator):
    def __init__(self, mu=0, sigma=1, lb=-1., ub=1., p=20):
        super(DataGeneratorScena3, self).__init__(mu, sigma, lb, ub, p)

    def get_context(self):
        return np.random.uniform(self._lb, self._ub, self.p)

    def get_reward(self, context, action):
        return (action > 0.5) * 10 * np.log(context[0] + 2) * max(action ** 2 - 0.25, 0) + np.random.normal(self._mu, self._sigma, 1)[0]


    def get_mean_reward(self, context, action):
        return (action > 0.5) * 10 * np.log(context[0] + 2) * max(action ** 2 - 0.25, 0)


class DataGeneratorScena4(DataGenerator):
    def __init__(self, mu=0, sigma=1, lb=-1., ub=1., p=20):
        super(DataGeneratorScena4, self).__init__(mu, sigma, lb, ub, p)

    def get_context(self):
        return np.random.uniform(self._lb, self._ub, self.p)

    def get_reward(self, context, action):
        return (8 + 4 * context[0] - 2 * context[1] - 2 * context[2] - 10 * (1 + 0.5 * context[0] + 0.5 * context[1] - 2 * action) ** 2) / 5 + np.random.normal(self._mu, self._sigma, 1)[0]

    def get_mean_reward(self, context, action):
        return (8 + 4 * context[0] - 2 * context[1] - 2 * context[2] - 10 * (1 + 0.5 * context[0] + 0.5 * context[1] - 2 * action) ** 2) / 5


class RealDataGenerator(DataGenerator):
    def __init__(self, file_name='real_envir.pickle'):
        with open(file_name, 'rb') as handle:
            envir= pickle.load(handle)

        self.org_data = envir[0]
        self.regr_mean = envir[1]
        self.sigma = envir[2]
        self.Q_nn = envir[3]
        self.tau = envir[4]

    def get_sample(self, policy):
        idx = random.sample(range(len(self.org_data)), 1)[0]
        context = self.org_data['xt'][idx]
        if policy == 'behavior':
            action = self.org_data['at'][idx]
        else:
            action = policy(context)

        # mean of y
        y_mean = self.regr_mean.predict(np.append(np.array(context),np.array(action)).reshape(1, len(context)+1))
        # std of y
        #res_mean = self.regr_res.predict(np.append(np.array(context),np.array(action)).reshape(1, len(context)+1))

        return context, action, np.random.normal(y_mean[0], self.sigma, 1)[0]

    def get_mean_reward(self, policy):
        idx = random.sample(range(len(self.org_data)), 1)[0]
        context = self.org_data['xt'][idx]
        if policy == 'behavior':
            action = self.org_data['at'][idx]
        else:
            action = policy(context)

        # mean of y
        y_mean = self.regr_mean.predict(np.append(np.array(context),np.array(action)).reshape(1, len(context)+1))

        return y_mean[0]


class ToyExample(DataGenerator):
    def __init__(self, mu=0, sigma=1, lb=0., ub=1., p=20):
        super(ToyExample, self).__init__(mu, sigma, lb, ub, p)

    def get_context(self):
        return np.random.uniform(self._lb, self._ub, self.p)[0]

    def get_reward(self, context, action):
        return (action > 0.5) * 10 * np.log(context + 2) * max(action ** 2 - 0.25, 0) + np.random.normal(self._mu, self._sigma, 1)[0]

    def get_mean_reward(self, context, action):
        return (action > 0.5) * 10 * np.log(context + 2) * np.maximum(action ** 2 - 0.25, 0)