-
Notifications
You must be signed in to change notification settings - Fork 1
/
space_declarations.py
96 lines (89 loc) · 6.54 KB
/
space_declarations.py
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
from hyperopt import hp
'''
Declare search space for hyperopt. Imported by hyperopt_search.py.
'''
n_nodes_options = [2 ** i for i in range(2, 12)] #range(2, 12)
activation_options = ['relu', 'sigmoid']
mlp_space = {
'n_nodes_layer1': hp.choice('n_nodes_layer1', n_nodes_options),
'layer1_activation': hp.choice('layer1_activation', activation_options),
'num_layers': hp.choice('num_layers', [{'layers': 'one'},
{'layers': 'two',
'nodes2': hp.choice('nodes2.2', n_nodes_options),
'activation2': hp.choice('activation2.2', activation_options)},
{'layers': 'three',
'nodes2': hp.choice('nodes3.2', n_nodes_options),
'activation2': hp.choice('activation3.2', activation_options),
'nodes3': hp.choice('nodes3.3', n_nodes_options),
'activation3': hp.choice('activation3.3', activation_options)},
{'layers': 'four',
'nodes2': hp.choice('nodes4.2', n_nodes_options),
'activation2': hp.choice('activation4.2', activation_options),
'nodes3': hp.choice('nodes4.3', n_nodes_options),
'activation3': hp.choice('activation4.3', activation_options),
'nodes4': hp.choice('nodes4.4', n_nodes_options),
'activation4': hp.choice('activation4.4', activation_options)},
{'layers': 'five',
'nodes2': hp.choice('nodes5.2', n_nodes_options),
'activation2': hp.choice('activation5.2', activation_options),
'nodes3': hp.choice('nodes5.3', n_nodes_options),
'activation3': hp.choice('activation5.3', activation_options),
'nodes4': hp.choice('nodes5.4', n_nodes_options),
'activation4': hp.choice('activation5.4', activation_options),
'nodes5': hp.choice('nodes5.5', n_nodes_options),
'activation5': hp.choice('activation5.5', activation_options)}
]),
'learning_rate': hp.choice('learning_rate', [0.01, 0.001, 0.0001, 0.00001]),
'batch_size': hp.choice('batch_size', [2 ** i for i in range(2, 7)]),
}
n_nodes_options = [2 ** i for i in range(2, 8)]
activation_options = ['relu', 'sigmoid']
fused_space = {
'n_nodes_layer1': hp.choice('n_nodes_layer1', n_nodes_options),
'layer1_activation': hp.choice('layer1_activation', activation_options),
'num_layers': hp.choice('num_layers', [{'layers': 'one'},
{'layers': 'two',
'nodes2': hp.choice('nodes2.2', n_nodes_options),
'activation2': hp.choice('activation2.2', activation_options)}
]),
'learning_rate': hp.choice('learning_rate', [0.01, 0.001, 0.0001, 0.00001]),
'loss': hp.choice('loss', ['mae','mse']),
'batch_size': hp.choice('batch_size', [2 ** i for i in range(3, 7)]),
'l2_2': hp.choice('l2_2', [ 0.1, 0.01, 0.001, 0.0001,0.00001, 0]),
'l2_1': hp.choice('l2_1', [ 0.1, 0.01, 0.001, 0.0001,0.00001, 0]),
'np_seed': hp.choice('np_seed', [0,1,2]),
'tf_seed': hp.choice('tf_seed', [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20])
}
n_convs_options = [16,32,64,128]
kernal_sz_options = [(2,2),(4,4),(8,8)]
cnn_space = {
'n_nodes_layer1': hp.choice('n_convs_layer1', n_convs_options),
'num_layers': hp.choice('num_layers', [{'layers': 'one',
'n_convs1': hp.choice('n_convs1.1', n_convs_options),
'kernal_sz1': hp.choice('kernal_sz1.2', kernal_sz_options)},
{'layers': 'two',
'n_convs1': hp.choice('n_convs2.1', n_convs_options),
'n_convs2': hp.choice('n_convs2.2', n_convs_options),
'kernal_sz1': hp.choice('kernal_sz2.1', kernal_sz_options),
'kernal_sz2': hp.choice('kernal_sz2.2', kernal_sz_options)},
{'layers': 'three',
'n_convs1': hp.choice('n_convs3.1', n_convs_options),
'n_convs2': hp.choice('n_convs3.2', n_convs_options),
'n_convs3': hp.choice('n_convs3.3', n_convs_options),
'kernal_sz1': hp.choice('kernal_sz3.1', kernal_sz_options),
'kernal_sz2': hp.choice('kernal_sz3.2', kernal_sz_options),
'kernal_sz3': hp.choice('kernal_sz3.3', kernal_sz_options)},
{'layers': 'four',
'n_convs1': hp.choice('n_convs4.1', n_convs_options),
'n_convs2': hp.choice('n_convs4.2', n_convs_options),
'n_convs3': hp.choice('n_convs4.3', n_convs_options),
'n_convs4': hp.choice('n_convs4.4', n_convs_options),
'kernal_sz1': hp.choice('kernal_sz4.1', kernal_sz_options),
'kernal_sz2': hp.choice('kernal_sz4.2', kernal_sz_options),
'kernal_sz3': hp.choice('kernal_sz4.3', kernal_sz_options),
'kernal_sz4': hp.choice('kernal_sz4.4', kernal_sz_options)}
]),
'batch_norm': hp.choice('batch_norm', [True, False]),
# 'pooling': hp.choice('pooling', [True, False]),
'learning_rate': hp.choice('learning_rate', [0.01, 0.001, 0.0001])
}