|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import numbers |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
import pandas as pd |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
class Converter: |
|
11
|
|
|
def __init__(self, search_space): |
|
12
|
|
|
self.search_space = search_space |
|
13
|
|
|
self.para_names = list(self.search_space.keys()) |
|
14
|
|
|
|
|
15
|
|
|
def value2position(self, value): |
|
16
|
|
|
position = [] |
|
17
|
|
|
for n, space_dim in enumerate(self.search_space_values): |
|
18
|
|
|
pos = np.abs(value[n] - space_dim).argmin() |
|
19
|
|
|
position.append(pos) |
|
20
|
|
|
|
|
21
|
|
|
return np.array(position).astype(int) |
|
22
|
|
|
|
|
23
|
|
|
def value2para(self, value): |
|
24
|
|
|
para = {} |
|
25
|
|
|
for key, p_ in zip(self.para_names, value): |
|
26
|
|
|
para[key] = p_ |
|
27
|
|
|
|
|
28
|
|
|
return para |
|
29
|
|
|
|
|
30
|
|
|
def position2value(self, position): |
|
31
|
|
|
value = [] |
|
32
|
|
|
|
|
33
|
|
|
for n, space_dim in enumerate(self.search_space_values): |
|
34
|
|
|
value.append(space_dim[position[n]]) |
|
35
|
|
|
|
|
36
|
|
|
return np.array(value) |
|
37
|
|
|
|
|
38
|
|
|
def positions2values(self, positions): |
|
39
|
|
|
values_temp = [] |
|
40
|
|
|
positions_np = np.array(positions) |
|
41
|
|
|
|
|
42
|
|
|
for n, space_dim in enumerate(self.search_space_values): |
|
43
|
|
|
pos_1d = positions_np[:, n] |
|
44
|
|
|
value_ = np.take(space_dim, pos_1d, axis=0) |
|
45
|
|
|
values_temp.append(value_) |
|
46
|
|
|
|
|
47
|
|
|
values = list(np.array(values_temp).T) |
|
48
|
|
|
return values |
|
49
|
|
|
|
|
50
|
|
|
def para2value(self, para): |
|
51
|
|
|
value = [] |
|
52
|
|
|
for para_name in self.para_names: |
|
53
|
|
|
value.append(para[para_name]) |
|
54
|
|
|
|
|
55
|
|
|
return np.array(value) |
|
56
|
|
|
|
|
57
|
|
|
def _memory2dataframe(self, memory_dict): |
|
58
|
|
|
positions = [np.array(pos).astype(int) for pos in list(memory_dict.keys())] |
|
59
|
|
|
scores = list(memory_dict.values()) |
|
60
|
|
|
|
|
61
|
|
|
memory_positions = pd.DataFrame(positions, columns=self.para_names) |
|
62
|
|
|
memory_positions["score"] = scores |
|
63
|
|
|
|
|
64
|
|
|
return memory_positions |
|
65
|
|
|
|
|
66
|
|
|
|
|
67
|
|
|
class HyperGradientTrafo(Converter): |
|
68
|
|
|
def __init__(self, search_space): |
|
69
|
|
|
super().__init__(search_space) |
|
70
|
|
|
self.search_space_values = list(self.search_space.values()) |
|
71
|
|
|
|
|
72
|
|
|
search_space_positions = {} |
|
73
|
|
|
for key in search_space.keys(): |
|
74
|
|
|
search_space_positions[key] = np.array(range(len(search_space[key]))) |
|
75
|
|
|
self.search_space_positions = search_space_positions |
|
76
|
|
|
|
|
77
|
|
|
""" |
|
78
|
|
|
self.search_space_ltm = {} |
|
79
|
|
|
self.data_types = {} |
|
80
|
|
|
for para_name in search_space.keys(): |
|
81
|
|
|
value0 = search_space[para_name][0] |
|
82
|
|
|
|
|
83
|
|
|
if isinstance(value0, numbers.Number): |
|
84
|
|
|
type0 = "number" |
|
85
|
|
|
search_dim_ltm = search_space[para_name] |
|
86
|
|
|
elif isinstance(value0, str): |
|
87
|
|
|
type0 = "string" |
|
88
|
|
|
search_dim_ltm = search_space[para_name] |
|
89
|
|
|
|
|
90
|
|
|
elif callable(value0): |
|
91
|
|
|
type0 = "function" |
|
92
|
|
|
|
|
93
|
|
|
search_dim_ltm = [] |
|
94
|
|
|
for func in list(search_space[para_name]): |
|
95
|
|
|
search_dim_ltm.append(func.__name__) |
|
96
|
|
|
|
|
97
|
|
|
else: |
|
98
|
|
|
type0 = None |
|
99
|
|
|
search_dim_ltm = search_space[para_name] |
|
100
|
|
|
|
|
101
|
|
|
self.data_types[para_name] = type0 |
|
102
|
|
|
self.search_space_ltm[para_name] = search_dim_ltm |
|
103
|
|
|
""" |
|
104
|
|
|
|
|
105
|
|
|
def trafo_initialize(self, initialize): |
|
106
|
|
|
if "warm_start" in list(initialize.keys()): |
|
107
|
|
|
warm_start = initialize["warm_start"] |
|
108
|
|
|
warm_start_gfo = [] |
|
109
|
|
|
for warm_start_ in warm_start: |
|
110
|
|
|
value = self.para2value(warm_start_) |
|
111
|
|
|
position = self.value2position(value) |
|
112
|
|
|
pos_para = self.value2para(position) |
|
113
|
|
|
|
|
114
|
|
|
warm_start_gfo.append(pos_para) |
|
115
|
|
|
|
|
116
|
|
|
initialize["warm_start"] = warm_start_gfo |
|
117
|
|
|
|
|
118
|
|
|
return initialize |
|
119
|
|
|
|
|
120
|
|
|
def get_list_positions(self, list1_values, search_dim): |
|
121
|
|
|
list_positions = [] |
|
122
|
|
|
|
|
123
|
|
|
for value2 in list1_values: |
|
124
|
|
|
pos_appended = False |
|
125
|
|
|
for value1 in search_dim: |
|
126
|
|
|
if value1 == value2: |
|
127
|
|
|
list_positions.append(search_dim.index(value1)) |
|
128
|
|
|
pos_appended = True |
|
129
|
|
|
break |
|
130
|
|
|
|
|
131
|
|
|
if not pos_appended: |
|
132
|
|
|
list_positions.append(None) |
|
133
|
|
|
|
|
134
|
|
|
return list_positions |
|
135
|
|
|
|
|
136
|
|
|
def trafo_memory_warm_start(self, results): |
|
137
|
|
|
if results is None: |
|
138
|
|
|
return results |
|
139
|
|
|
|
|
140
|
|
|
df_positions_dict = {} |
|
141
|
|
|
for para_name in self.para_names: |
|
142
|
|
|
result_dim_values = list(results[para_name].values) |
|
143
|
|
|
search_dim = self.search_space[para_name] |
|
144
|
|
|
|
|
145
|
|
|
# if self.data_types[para_name] == "function": |
|
146
|
|
|
# result_dim_values = [value.__name__ for value in result_dim_values] |
|
147
|
|
|
|
|
148
|
|
|
# print("\n para_name", para_name) |
|
149
|
|
|
|
|
150
|
|
|
# print(" result_dim_values", result_dim_values) |
|
151
|
|
|
# print(" search_dim", search_dim) |
|
152
|
|
|
|
|
153
|
|
|
list1_positions = self.get_list_positions(result_dim_values, search_dim) |
|
154
|
|
|
|
|
155
|
|
|
# remove None |
|
156
|
|
|
# list1_positions_ = [x for x in list1_positions if x is not None] |
|
157
|
|
|
df_positions_dict[para_name] = list1_positions |
|
158
|
|
|
|
|
159
|
|
|
results_new = pd.DataFrame(df_positions_dict) |
|
160
|
|
|
results_new["score"] = results["score"] |
|
161
|
|
|
results_new.dropna(how="any", inplace=True) |
|
162
|
|
|
|
|
163
|
|
|
return results_new |
|
164
|
|
|
|
SimonBlanke /
Hyperactive
