StateFeaturesLearning/environments.py at master · Sephora-M/StateFeaturesLearning · 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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
from learning_maze import LearningMazeDomain
from lspi import domains, basis_functions, policy, lspi, solvers

def example_grid_maze():
    height = 10
    width = 10
    reward_location = 9
    initial_state = None  # np.array([25])
    obstacles_location = [14, 13, 24, 23, 29, 28, 39, 38]  # range(height*width)
    walls_location = [50, 51, 52, 53, 54, 55, 56, 74, 75, 76, 77, 78, 79]
    obstacles_transition_probability = .2
    domain = domains.GridMazeDomain(height, width, reward_location,
                                         walls_location, obstacles_location, initial_state,
                                         obstacles_transition_probability)
    maze = LearningMazeDomain(domain=domain, num_sample=2000)

    return maze


def low_stretch_tree_maze(num_sample=100, length_sample=100):
    reward_location = [15]
    obstacles_location = []
    obstacles_transition_probability = .2
    domain = domains.SymmetricMazeDomain(rewards_locations=reward_location,
                                              obstacles_location=obstacles_location)
    maze = LearningMazeDomain(domain=domain, num_sample=num_sample, length_sample=length_sample)

    return maze


def tworooms(num_sample=100, length_sample=100, discount=0.9):
    height = 10
    width = 10
    reward_location = 18
    initial_state = None  # np.array([25])
    obstacles_location = []  # range(height*width)
    walls_location = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
                      10, 20, 30, 40, 50, 60, 70, 80, 90,
                      9, 19, 29, 39, 49, 59, 69, 79, 89, 99,
                      90, 91, 92, 93, 94, 95, 96, 97, 98, 99,
                      41, 42, 43, 44, 46, 47, 48, 49]
    obstacles_transition_probability = .2
    domain = domains.GridMazeDomain(height, width, reward_location,
                                         walls_location, obstacles_location, initial_state,
                                         obstacles_transition_probability)
    maze = LearningMazeDomain(domain=domain, num_sample=num_sample, length_sample=length_sample, discount=discount)

    return maze


def threerooms(num_sample=5000, length_sample=100):
    height = 50
    width = 100
    reward_location = 198
    initial_state = None  # np.array([25])
    obstacles_location = []  # range(height*width)
    walls_location = []
    walls_location.extend(range(100))
    walls_location.extend(range(4900, 5000))
    walls_location.extend(range(0, 5000, 100))
    walls_location.extend(range(99, 5000, 100))
    walls_location.extend(range(1600, 1670))
    walls_location.extend(range(1680, 1700))
    walls_location.extend(range(3200, 3220))
    walls_location.extend(range(3230, 3300))

    obstacles_transition_probability = .2
    domain = domains.GridMazeDomain(height, width, reward_location,
                                         walls_location, obstacles_location, initial_state,
                                         obstacles_transition_probability)
    maze = LearningMazeDomain(domain=domain, num_sample=num_sample, length_sample=length_sample)


    return maze


class FourRooms():

    def __init__(self, num_samples, length_sample, discount, initial_state=None):
        self.num_actions = 4
        self.height = 13
        self.width = 13
        self.initial_state = initial_state  # np.array([25])
        self.obstacles_location = []  # range(height*width)
        self.length_sample = length_sample
        self.num_samples = num_samples
        self.discount = discount
        self.sampling_policy = policy.Policy(basis_functions.FakeBasis(4), discount, 1)
        self.samples = []
        self.lspi_samples = []
        self.walks = []
        self.solver = solvers.LSTDQSolver()

        walls_location = []
        walls_location.extend(range(13))
        walls_location.extend(range(156, 169))
        walls_location.extend(range(0, 157, 13))
        walls_location.extend(range(12, 169, 13))
        walls_location.extend(range(78, 80))
        walls_location.extend(range(81, 85))
        walls_location.extend(range(97, 100))
        walls_location.extend(range(101, 104))
        walls_location.extend(range(6, 45, 13))
        walls_location.extend(range(58, 124, 13))
        walls_location.extend(range(149, 163, 13))
        self.walls_location = walls_location

        self.obstacles_transition_probability = .2

    def learn_n2v_basis(self, dimension, discount, epochs, window_size, learning_rate=0.5):
        basis = basis_functions.Node2vecBasis('',
            self.num_actions, self.domain.transition_probabilities, dimension, self.walks, epochs=epochs,
            window_size=window_size, workers=8)
        return basis

    def learn_basis(self, dimension, discount, epochs, window_size, learning_rate=0.5):
        basis = basis_functions.DiscountedNode2vecBasis(
            self.num_actions, self.domain.transition_probabilities, discount, dimension, self.walks, self.actions, epochs=epochs,
            window_size=window_size, learning_rate=learning_rate, workers=8)
        return basis

    def update_domain(self, new_reward_location, obstacles_location, num_samples, length_sample):
        self.reward_location = new_reward_location
        self.obstacles_location = obstacles_location
        self.num_samples = num_samples
        self.length_sample = length_sample
        self.domain = domains.GridMazeDomain(self.height, self.width, new_reward_location,
                                             self.walls_location, self.obstacles_location, self.initial_state,
                                             self.obstacles_transition_probability)
        self.compute_samples()

    def compute_samples(self, reset_samples=True, reset_policy=True, biased_walk=False):
        if reset_policy:
            self.sampling_policy = policy.Policy(basis_functions.FakeBasis(4), self.discount, 1)
        if reset_samples:
            self.samples = []
            self.lspi_samples = []
            self.walks = []
            self.actions = []
        for i in range(self.num_samples):
            if biased_walk:
                sample, walk, terminated, lspi_sample = self.domain.generate_unique_samples(self.length_sample,
                                                                                            self.sampling_policy)
            else:
                sample, walk, walk_actions, terminated, lspi_sample = self.domain.generate_samples(self.length_sample,
                                                                                     self.sampling_policy)
            self.samples.extend(sample)
            self.walks.append(walk)
            self.actions.append(walk_actions)
            # if terminated: # and len(self.lspi_samples) <= NUM_SAMPLES:
            self.lspi_samples.extend(lspi_sample)
        # print(self.lspi_samples)

    def run_lspi(self, basis, discount, max_iter, explore):
        basis_policy = policy.Policy(basis, discount, explore)
        learned_policy, distances, iterations = lspi.learn(self.lspi_samples, basis_policy, self.solver,
                                                           max_iterations=max_iter)

        return learned_policy, distances, iterations


def oneroom(plotV=True, num_sample=100, length_sample=100, computeV=False):
    height = 10
    width = 10
    reward_location = 9
    initial_state = None  # np.array([25])
    obstacles_location = []  # range(height*width)
    walls_location = []
    obstacles_transition_probability = .2
    domain = domains.GridMazeDomain(height, width, reward_location,
                                         walls_location, obstacles_location, initial_state,
                                         obstacles_transition_probability)
    maze = LearningMazeDomain(domain=domain, num_sample=num_sample, length_sample=length_sample)

    return maze


def obstacles_room(plotV=True, num_sample=100, length_sample=100, computeV=False):
    height = 10
    width = 10
    reward_location = 18
    initial_state = None  # np.array([25])
    obstacles_location = [12, 13, 22, 23,
                          35, 36, 45, 46,
                          62, 63, 72, 73,
                          67, 77]
    walls_location = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
                      10, 20, 30, 40, 50, 60, 70, 80, 90,
                      9, 19, 29, 39, 49, 59, 69, 79, 89, 99,
                      90, 91, 92, 93, 94, 95, 96, 97, 98, 99]
    obstacles_transition_probability = .2
    domain = domains.GridMazeDomain(height, width, reward_location,
                                         walls_location, obstacles_location, initial_state,
                                         obstacles_transition_probability)
    maze = LearningMazeDomain(domain=domain, num_sample=num_sample, length_sample=length_sample)

    return maze