MetaQNN/State_enumerator.py at main · hardikk65/MetaQNN · 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

import state_space_parameters as ssp
from helper import *

'''
    Types of Layers:
    1. Conv (Conv)
    2. Pooling (Pool)
    3. Fully Connected (fc)
    4. Global Average Pooling (GAP)
    5. Softmax (SM)


    Attributes:
    layer_depth < 12
    Representation_Size -----> {n >= 8, 4 <= n < 8 ,n < 4}

    kernel_size ---> {1,3,5}
    strides ----> {1 (Conv), 2(Pool)}
    No. of Channels ----> {64,128,256,512} (Conv), {Prev_layer channels} (Pooling)


    Consecutive_layers -----> {n < 3}
    Neurons -------> {512,256,128}


    s ---->previous_state
    Global_average_pooling/Softmax -----> Termination_State

'''


# TODO resolve no transition error


class State:

    def __init__(self,
            layer_type:str,
            kernel_size:int,
            channels:int,
            strides:int,
            image_size:int,
            layer_depth:int,
            neurons:int,
            fc_layers:int):


        self.layer_type = layer_type
        self.kernel_size = kernel_size
        self.channels = channels
        self.strides = strides
        self.image_size = image_size
        self.layer_depth = layer_depth
        self.neurons = neurons
        self.fc_layers = fc_layers

    def __repr__(self) -> str:
        return f"layer_type: {self.layer_type},kernel_size: {self.kernel_size},channels: {self.channels},strides: {self.strides},image_size: {self.image_size},layer_depth: {self.layer_depth},neurons: {self.neurons},fc_layers: {self.fc_layers}"


'''

    Possible Transitions:

    Conv ----> Conv,Pool,Fc
    Pool ----> Conv,Fc
    Fc   ----> Fc,SM


'''

class transitions:

    def __init__(self,A:State):
        self.A = A
        self.layer_type = A.layer_type
        self.kernel_size = A.kernel_size
        self.channels = A.channels
        self.strides = A.strides

        self.image_size = A.image_size
        if(self.layer_type == 'Conv' or self.layer_type == 'Pool'):
            self.image_size = calculate_image_size(self.image_size,self.kernel_size,self.strides)

        self.layer_depth = A.layer_depth
        self.neurons = A.neurons
        self.fc_layers = A.fc_layers

        self._data = load_data()


    def possible_transitions(self):

        all_transitions = []
        q_values = []
        if self.layer_depth == 11 or self.fc_layers == 2:
            all_transitions.append(
                State("SM",kernel_size=-1,channels=-1,strides=-1,image_size=-1,layer_depth=self.layer_depth + 1,neurons=0,fc_layers=self.fc_layers)
                )


        else:

            if self.layer_type == 'Conv':

                # Conv to Conv Transition
                if pool_size(self.image_size) >= 8 and self.layer_depth < 10:

                    for kernels in ssp.kernel_size:
                        for channels in ssp.Channels:
                            all_transitions.append(
                                State(layer_type="Conv",kernel_size=kernels,channels=channels,strides=1,image_size=pool_size(self.image_size),layer_depth=self.layer_depth + 1,neurons=self.neurons,fc_layers=self.fc_layers)
                            )

                    # Conv to Pool Transition
                    for kernels in ssp.kernel_size:
                        all_transitions.append(
                            State(layer_type="Pool",kernel_size=kernels,channels=self.channels,strides = 2,image_size=pool_size(self.image_size),layer_depth=self.layer_depth + 1,neurons=self.neurons,fc_layers = self.fc_layers)
                        )

                # Conv to fc Transition

                if (self.fc_layers < 2) or (pool_size(self.image_size) < 8 and pool_size(self.image_size) >= 4):
                    for neurons in ssp.neurons:

                        if self.neurons <= neurons:
                            all_transitions.append(
                                State(layer_type="fc",kernel_size=-1,channels = -1,strides = -1,image_size=self.image_size,layer_depth=self.layer_depth + 1,neurons = neurons,fc_layers=self.fc_layers + 1)
                            )


            elif self.layer_type == "Pool":

                # Pool to Conv Transition

                if self.layer_depth < 10:
                    for kernels in ssp.kernel_size:
                        for channels in ssp.Channels:
                            all_transitions.append(
                                State(layer_type="Conv",kernel_size=kernels,channels=channels,strides=1,image_size=pool_size(self.image_size),layer_depth=self.layer_depth + 1,neurons=self.neurons,fc_layers=self.fc_layers)
                            )


                # Pool to fc Transition
                if (self.fc_layers < 2) and (pool_size(self.image_size) < 8 and pool_size(self.image_size) >= 4):
                    for neurons in ssp.neurons:
                        if self.neurons <= neurons:
                            all_transitions.append(
                                State(layer_type="fc",kernel_size=-1,channels = -1,strides = -1,image_size=self.image_size,layer_depth=self.layer_depth + 1,neurons = neurons,fc_layers=self.fc_layers + 1,)
                            )


            elif self.layer_type == 'fc':

                if self.fc_layers < 2:
                    for neurons in ssp.neurons:
                        if self.neurons <= neurons:
                            all_transitions.append(
                                State(layer_type="fc",kernel_size=-1,channels = -1,strides = -1,image_size=self.image_size,layer_depth=self.layer_depth + 1,neurons = neurons,fc_layers=self.fc_layers + 1,)
                            )

                all_transitions.append(
                State("SM",kernel_size=-1,channels=-1,strides=-1,image_size=-1,layer_depth=self.layer_depth + 1,neurons=0,fc_layers=0)
                )


        for states in all_transitions:
            for end_state,utils in zip(self._data.loc[self._data['Start-State'] == self.A.__repr__()]['End-State'],self._data.loc[self._data['Start-State'] == self.A.__repr__()]['Utility']):
                if end_state == states.__repr__():
                    q_values.append(utils)

        return all_transitions,q_values