File size: 4,957 Bytes
52933b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch

from .nets.attention_model.attention_model import *


class Problem:
    def __init__(self, name):
        self.NAME = name


class Backbone(nn.Module):
    def __init__(
        self,
        embedding_dim=128,
        problem_name="tsp",
        n_encode_layers=3,
        tanh_clipping=10.0,
        n_heads=8,
        device="cpu",
    ):
        super(Backbone, self).__init__()
        self.device = device
        self.problem = Problem(problem_name)
        self.embedding = AutoEmbedding(self.problem.NAME, {"embedding_dim": embedding_dim})

        self.encoder = GraphAttentionEncoder(
            n_heads=n_heads,
            embed_dim=embedding_dim,
            n_layers=n_encode_layers,
        )

        self.decoder = Decoder(
            embedding_dim, self.embedding.context_dim, n_heads, self.problem, tanh_clipping
        )

    def forward(self, obs):
        state = stateWrapper(obs, device=self.device, problem=self.problem.NAME)
        input = state.states["observations"]
        embedding = self.embedding(input)
        encoded_inputs, _ = self.encoder(embedding)

        # decoding
        cached_embeddings = self.decoder._precompute(encoded_inputs)
        logits, glimpse = self.decoder.advance(cached_embeddings, state)

        return logits, glimpse

    def encode(self, obs):
        state = stateWrapper(obs, device=self.device, problem=self.problem.NAME)
        input = state.states["observations"]
        embedding = self.embedding(input)
        encoded_inputs, _ = self.encoder(embedding)
        cached_embeddings = self.decoder._precompute(encoded_inputs)
        return cached_embeddings

    def decode(self, obs, cached_embeddings):
        state = stateWrapper(obs, device=self.device, problem=self.problem.NAME)
        logits, glimpse = self.decoder.advance(cached_embeddings, state)

        return logits, glimpse


class Actor(nn.Module):
    def __init__(self):
        super(Actor, self).__init__()

    def forward(self, x):
        logits = x[0]  # .squeeze(1) # not needed for pomo
        return logits


class Critic(nn.Module):
    def __init__(self, *args, **kwargs):
        super(Critic, self).__init__()
        hidden_size = kwargs["hidden_size"]
        self.mlp = nn.Sequential(
            nn.Linear(hidden_size, hidden_size), nn.ReLU(), nn.Linear(hidden_size, 1)
        )

    def forward(self, x):
        out = self.mlp(x[1])  # B x T x h_dim --mlp--> B x T X 1
        return out


class Agent(nn.Module):
    def __init__(self, embedding_dim=128, device="cpu", name="tsp"):
        super().__init__()
        self.backbone = Backbone(embedding_dim=embedding_dim, device=device, problem_name=name)
        self.critic = Critic(hidden_size=embedding_dim)
        self.actor = Actor()

    def forward(self, x):  # only actor
        x = self.backbone(x)
        logits = self.actor(x)
        action = logits.max(2)[1]
        return action, logits

    def get_value(self, x):
        x = self.backbone(x)
        return self.critic(x)

    def get_action_and_value(self, x, action=None):
        x = self.backbone(x)
        logits = self.actor(x)
        probs = torch.distributions.Categorical(logits=logits)
        if action is None:
            action = probs.sample()
        return action, probs.log_prob(action), probs.entropy(), self.critic(x)

    def get_value_cached(self, x, state):
        x = self.backbone.decode(x, state)
        return self.critic(x)

    def get_action_and_value_cached(self, x, action=None, state=None):
        if state is None:
            state = self.backbone.encode(x)
            x = self.backbone.decode(x, state)
        else:
            x = self.backbone.decode(x, state)
        logits = self.actor(x)
        probs = torch.distributions.Categorical(logits=logits)
        if action is None:
            action = probs.sample()
        return action, probs.log_prob(action), probs.entropy(), self.critic(x), state


class stateWrapper:
    """
    from dict of numpy arrays to an object that supplies function and data
    """

    def __init__(self, states, device, problem="tsp"):
        self.device = device
        self.states = {k: torch.tensor(v, device=self.device) for k, v in states.items()}
        if problem == "tsp":
            self.is_initial_action = self.states["is_initial_action"].to(torch.bool)
            self.first_a = self.states["first_node_idx"]
        elif problem == "cvrp":
            input = {
                "loc": self.states["observations"],
                "depot": self.states["depot"].squeeze(-1),
                "demand": self.states["demand"],
            }
            self.states["observations"] = input
            self.VEHICLE_CAPACITY = 0
            self.used_capacity = -self.states["current_load"]

    def get_current_node(self):
        return self.states["last_node_idx"]

    def get_mask(self):
        return (1 - self.states["action_mask"]).to(torch.bool)