GameModel/defender.py at master · wyz2368/GameModel · 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
import random
import numpy as np
import sample_strategy as ss

class Defender(object):

    def __init__(self, G):
        self.num_nodes = G.number_of_nodes()
        self.observation = []
        self.history = 3
        self.prev_obs = [0]*self.num_nodes*(self.history - 1)

        self.defact = set()
        self.prev_defact = [set()]*self.history

        self.rand_limit = 4


    #TODO: nn should input mask!!!!!!! ALL zeros.
    def def_greedy_action_builder(self, G, timeleft):
        self.defact.clear()
        isDup = False
        mask = np.zeros(shape=(1, self.num_nodes), dtype=np.float32)
        #TODO: sample a strategy
        nn = ss.sample_strategy_from_mixed(env=self.myenv,str_set=self.str_set,mix_str=self.mix_str,identity=0)
        self.set_current_strategy(nn)
        while not isDup:
            def_input = self.def_obs_constructor(G, timeleft)
            x = self.nn_def(def_input[None], mask)[0] #corrensponding to baselines
            if not isinstance(x,int):
                raise ValueError("The chosen action is not an integer.")
            action_space = self.get_def_actionspace(G)
            action = action_space[x] #TODO: make sure whether x starting from 0 or 1.
            if action == 'pass':
                break
            isDup = (action in self.defact)
            if not isDup:
                self.defact.add(action)

    def def_greedy_action_builder_single(self, G, timeleft, nn_def):
        self.defact.clear()
        isDup = False
        mask = np.zeros(shape=(1, self.num_nodes), dtype=np.float32)

        while not isDup:
            def_input = self.def_obs_constructor(G, timeleft)
            x = nn_def(def_input[None], mask)[0] #corrensponding to baselines
            if not isinstance(x,int):
                raise ValueError("The chosen action is not an integer.")
            action_space = self.get_def_actionspace(G)
            action = action_space[x] # x starting from 0.
            if action == 'pass':
                break
            isDup = (action in self.defact)
            if not isDup:
                self.defact.add(action)

    #TODO: Since initialize with zeros, clean these funcs.
    def def_obs_constructor(self, G, timeleft):
        wasdef = self.get_def_wasDefended(G)
        indef = self.get_def_inDefenseSet(G)
        def_input = self.prev_obs + self.observation + wasdef + indef + [timeleft]
        return np.array(def_input)

    # TODO: Can we do matrix operation to get this vector?
    def get_def_wasDefended(self, G):
        wasdef = []
        #old defact is added first.
        for obs in self.prev_defact:
            for node in G.nodes:
                if node in obs:
                    wasdef.append(1)
                else:
                    wasdef.append(0)
        return wasdef

    def get_def_inDefenseSet(self, G):
        indef = []
        for node in G.nodes:
            if node in self.defact:
                indef.append(1)
            else:
                indef.append(0)
        return indef

    def get_def_actionspace(self, G):
        actionspace = list(G.nodes) + ['pass']
        return actionspace

    def uniform_strategy(self, G):
        return set(sorted(random.sample(list(G.nodes), self.rand_limit)))

    def cut_prev_obs(self):
        if len(self.prev_obs)/self.num_nodes > self.history - 1:
            self.prev_obs = self.prev_obs[(- self.history + 1)*self.num_nodes:]

    def cut_prev_defact(self):
        if len(self.prev_defact) > self.history:
            self.prev_defact = self.prev_defact[-self.history:]

    def save_defact2prev(self):
        print("Don't forget to update defact after save_defact2prev.")
        self.prev_defact.append(self.defact)
        self.cut_prev_defact()

    def update_obs(self, obs):
        #TODO: check obs update for the first time step
        self.prev_obs += self.observation #TODO: prev_obs is a list. Do not append.
        self.cut_prev_obs()
        self.observation = obs
    # TODO: in env, feed noisy obs to the observation

    def update_history(self, history):
        self.history = history

    def reset_def(self):
        self.observation = []
        self.prev_obs = [0] * self.num_nodes * (self.history - 1)
        self.defact.clear()
        self.prev_defact = [set()] * self.history

    def set_current_strategy(self,strategy):
        self.nn_def = strategy

    #TODO: call this while creating env
    def set_env_belong_to(self,env):
        self.myenv = env

    def set_mix_strategy(self,mix):
        self.mix_str = mix

    def set_str_set(self,set):
        self.str_set = set