|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import os |
|
6
|
|
|
import json |
|
7
|
|
|
import dill |
|
8
|
|
|
import inspect |
|
9
|
|
|
|
|
10
|
|
|
import numpy as np |
|
11
|
|
|
import pandas as pd |
|
12
|
|
|
|
|
13
|
|
|
from .memory_io import MemoryIO |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
class MemoryDump(MemoryIO): |
|
17
|
|
|
def __init__(self, _space_, _main_args_, _cand_, memory_dict): |
|
18
|
|
|
super().__init__(_space_, _main_args_, _cand_, memory_dict) |
|
19
|
|
|
|
|
20
|
|
|
self.memory_type = _main_args_.memory |
|
21
|
|
|
self.memory_dict = memory_dict |
|
22
|
|
|
|
|
23
|
|
|
def _save_memory(self, _main_args_, _opt_args_, _cand_): |
|
24
|
|
|
path = self._get_file_path(_cand_.func_) |
|
25
|
|
|
meta_data = self._collect(_cand_) |
|
26
|
|
|
|
|
27
|
|
|
meta_data["run"] = self.datetime |
|
28
|
|
|
self._save_toCSV(meta_data, path) |
|
29
|
|
|
|
|
30
|
|
|
obj_func_path = self.func_path + "objective_function.py" |
|
31
|
|
|
if not os.path.exists(obj_func_path): |
|
32
|
|
|
file = open(obj_func_path, "w") |
|
33
|
|
|
file.write(self._get_func_str(_cand_.func_)) |
|
34
|
|
|
file.close() |
|
35
|
|
|
|
|
36
|
|
|
search_config_path = self.date_path + "search_config.py" |
|
37
|
|
|
search_config_temp = dict(self._main_args_.search_config) |
|
38
|
|
|
|
|
39
|
|
|
for key in search_config_temp.keys(): |
|
40
|
|
|
if isinstance(key, str): |
|
41
|
|
|
continue |
|
42
|
|
|
search_config_temp[key.__name__] = search_config_temp[key] |
|
43
|
|
|
del search_config_temp[key] |
|
44
|
|
|
|
|
45
|
|
|
search_config_str = "search_config = " + str(search_config_temp) |
|
46
|
|
|
|
|
47
|
|
|
if not os.path.exists(search_config_path): |
|
48
|
|
|
file = open(search_config_path, "w") |
|
49
|
|
|
file.write(search_config_str) |
|
50
|
|
|
file.close() |
|
51
|
|
|
|
|
52
|
|
|
""" |
|
53
|
|
|
os.chdir(self.date_path) |
|
54
|
|
|
os.system("black search_config.py") |
|
55
|
|
|
os.getcwd() |
|
56
|
|
|
""" |
|
57
|
|
|
|
|
58
|
|
|
run_data = { |
|
59
|
|
|
"random_state": self._main_args_.random_state, |
|
60
|
|
|
"max_time": self._main_args_.random_state, |
|
61
|
|
|
"n_iter": self._main_args_.n_iter, |
|
62
|
|
|
"optimizer": self._main_args_.optimizer, |
|
63
|
|
|
"n_jobs": self._main_args_.n_jobs, |
|
64
|
|
|
"eval_time": np.array(_cand_.eval_time).sum(), |
|
65
|
|
|
"total_time": _cand_.total_time, |
|
66
|
|
|
} |
|
67
|
|
|
|
|
68
|
|
|
with open(self.date_path + "run_data.json", "w") as f: |
|
69
|
|
|
json.dump(run_data, f, indent=4) |
|
70
|
|
|
|
|
71
|
|
|
""" |
|
72
|
|
|
print("_opt_args_.kwargs_opt", _opt_args_.kwargs_opt) |
|
73
|
|
|
|
|
74
|
|
|
opt_para = pd.DataFrame.from_dict(_opt_args_.kwargs_opt, dtype=object) |
|
75
|
|
|
print("opt_para", opt_para) |
|
76
|
|
|
opt_para.to_csv(self.date_path + "opt_para", index=False) |
|
77
|
|
|
""" |
|
78
|
|
|
|
|
79
|
|
|
def _get_func_str(self, func): |
|
80
|
|
|
return inspect.getsource(func) |
|
81
|
|
|
|
|
82
|
|
|
def _get_file_path(self, model_func): |
|
83
|
|
|
if not os.path.exists(self.date_path): |
|
84
|
|
|
os.makedirs(self.date_path) |
|
85
|
|
|
|
|
86
|
|
|
return self.func_path + (self.feature_hash + "_" + self.label_hash + "_.csv") |
|
87
|
|
|
|
|
88
|
|
|
def _collect(self, _cand_): |
|
89
|
|
|
results_dict = self._get_opt_meta_data() |
|
90
|
|
|
|
|
91
|
|
|
para_pd = pd.DataFrame(results_dict["params"]) |
|
92
|
|
|
metric_pd = pd.DataFrame( |
|
93
|
|
|
results_dict["mean_test_score"], columns=["mean_test_score"] |
|
94
|
|
|
) |
|
95
|
|
|
n_rows = len(para_pd) |
|
96
|
|
|
eval_time = pd.DataFrame(_cand_.eval_time[-n_rows:], columns=["eval_time"]) |
|
97
|
|
|
md_model = pd.concat( |
|
98
|
|
|
[para_pd, metric_pd, eval_time], axis=1, ignore_index=False |
|
99
|
|
|
) |
|
100
|
|
|
|
|
101
|
|
|
return md_model |
|
102
|
|
|
|
|
103
|
|
|
def _get_opt_meta_data(self): |
|
104
|
|
|
results_dict = {} |
|
105
|
|
|
para_list = [] |
|
106
|
|
|
score_list = [] |
|
107
|
|
|
|
|
108
|
|
|
for key in self.memory_dict.keys(): |
|
109
|
|
|
pos = np.fromstring(key, dtype=int) |
|
110
|
|
|
para = self._space_.pos2para(pos) |
|
111
|
|
|
score = self.memory_dict[key] |
|
112
|
|
|
|
|
113
|
|
|
for key in para.keys(): |
|
114
|
|
|
if ( |
|
115
|
|
|
not isinstance(para[key], int) |
|
116
|
|
|
and not isinstance(para[key], float) |
|
117
|
|
|
and not isinstance(para[key], str) |
|
118
|
|
|
): |
|
119
|
|
|
|
|
120
|
|
|
para_dill = dill.dumps(para[key]) |
|
121
|
|
|
para_hash = self._get_hash(para_dill) |
|
122
|
|
|
|
|
123
|
|
|
with open( |
|
124
|
|
|
self.func_path + str(para_hash) + ".pkl", "wb" |
|
125
|
|
|
) as pickle_file: |
|
126
|
|
|
dill.dump(para_dill, pickle_file) |
|
127
|
|
|
|
|
128
|
|
|
para[key] = para_hash |
|
129
|
|
|
|
|
130
|
|
|
if score != 0: |
|
131
|
|
|
para_list.append(para) |
|
132
|
|
|
score_list.append(score) |
|
133
|
|
|
|
|
134
|
|
|
results_dict["params"] = para_list |
|
135
|
|
|
results_dict["mean_test_score"] = score_list |
|
136
|
|
|
|
|
137
|
|
|
return results_dict |
|
138
|
|
|
|
|
139
|
|
|
def _save_toCSV(self, meta_data_new, path): |
|
140
|
|
|
if os.path.exists(path): |
|
141
|
|
|
meta_data_old = pd.read_csv(path) |
|
142
|
|
|
|
|
143
|
|
|
if len(meta_data_old.columns) != len(meta_data_new.columns): |
|
144
|
|
|
print("Warning meta data dimensionality does not match") |
|
145
|
|
|
print("Meta data will not be saved") |
|
146
|
|
|
return |
|
147
|
|
|
|
|
148
|
|
|
meta_data = meta_data_old.append(meta_data_new) |
|
149
|
|
|
|
|
150
|
|
|
columns = list(meta_data.columns) |
|
151
|
|
|
noScore = ["mean_test_score", "cv_default_score", "eval_time", "run"] |
|
152
|
|
|
columns_noScore = [c for c in columns if c not in noScore] |
|
153
|
|
|
|
|
154
|
|
|
meta_data = meta_data.drop_duplicates(subset=columns_noScore) |
|
155
|
|
|
else: |
|
156
|
|
|
meta_data = meta_data_new |
|
157
|
|
|
|
|
158
|
|
|
meta_data.to_csv(path, index=False) |
|
159
|
|
|
|
SimonBlanke /
Hyperactive
