This repository has been archived by the owner on Apr 22, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
optimize.py
93 lines (73 loc) · 2.76 KB
/
optimize.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
from run_model import run_model, calc_derivations
import numpy
from tqdm import tqdm
import os
default_params = {
'epochs': 100,
'layers': 1,
'neurons': 10,
'batch': 512,
'weight': 'he_normal',
'activation': 'relu',
'verbose': 0
}
opt_params = {
'epochs': 90,
'layers': 5,
'neurons': 60,
'batch': 300,
'weight': 'glorot_normal',
'activation': 'relu'
}
default_param_ranges = {
'epochs': list(range(1,10)) + list(range(10, 101, 10)),
'layers': list(range(1, 11)),
'neurons': list(range(1,10)) + list(range(10, 101, 10)),
'batch': list(range(100, 1000, 100)) + list(range(1000, 6001, 1000)),
'weight': ['he_normal', 'he_uniform', 'lecun_uniform', 'lecun_normal', 'glorot_normal', 'glorot_uniform'],
'activation': ['relu', 'selu', 'tanh', 'elu', 'sigmoid'],
}
def optimize(params, ranges, iterations, save_file):
optimal_params = params
os.makedirs(save_file)
for i in tqdm(range(iterations), desc='Optimization iterations'):
param_value_accuracies = experiment_ranges(optimal_params, ranges, save_file + '/' + str(i) + '.npy')
for param, (param_values, accuracies) in param_value_accuracies.items():
optimal_param_value = max_param(param_values, accuracies)
optimal_params[param] = optimal_param_value
return optimal_params
def max_param(param_values, accuracies):
return param_values[accuracies.index(max(accuracies))]
def make_param_ranges(params1, params2):
param_ranges = {}
for param in params1:
param_ranges[param] = [params1[param]]
for param in params2:
if param in param_ranges:
param_ranges[param].append(params2[param])
else:
param_ranges[param] = [params2[param]]
return param_ranges
def experiment_ranges(params, ranges, save_file):
param_value_accuracies = {}
for param, param_values in tqdm(ranges.items(), desc='Parameter ranges'):
print(param)
accuracies = experiment(params, param, param_values)
param_value_accuracies[param] = param_values, accuracies
numpy.save(save_file, {'params': params, 'param_value_accuracies': param_value_accuracies})
return param_value_accuracies
def experiment(params, param, param_values):
accuracies = []
new_params = {**default_params, **params}
for value in tqdm(param_values, desc='Parameter values'):
new_params[param] = value
print(value)
_, _, truth_values, predictions = run_model(new_params)
accuracy, _, _ = calc_derivations(truth_values, predictions)
accuracies.append(accuracy)
return accuracies
def main():
optimal_params = optimize(opt_params, default_param_ranges, 2, 'optimize1')
print(optimal_params)
if __name__ == "__main__":
main()