|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import os |
|
6
|
|
|
import dill |
|
7
|
|
|
import shutil |
|
8
|
|
|
import pandas as pd |
|
9
|
|
|
|
|
10
|
|
|
from .ltm_data_path import ltm_data_path |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
def merge_unhashable_df(df1, df2): |
|
14
|
|
|
columns = df1.columns |
|
15
|
|
|
columns0 = df2.columns |
|
16
|
|
|
|
|
17
|
|
|
if set(columns) != set(columns0): |
|
18
|
|
|
print("Error columns of df1 and df2 must be the same") |
|
19
|
|
|
return |
|
20
|
|
|
|
|
21
|
|
|
df1["item"] = range(len(df1)) |
|
22
|
|
|
df2["item"] = range(len(df1), len(df1) + len(df2)) |
|
23
|
|
|
|
|
24
|
|
|
result = pd.merge(df1, df2, on="item", how="outer", suffixes=["", "_y"]) |
|
25
|
|
|
|
|
26
|
|
|
for col in columns: |
|
27
|
|
|
result[col].update(result[col + "_y"]) |
|
28
|
|
|
|
|
29
|
|
|
result = result[result.columns[~result.columns.str.endswith("_y")]] |
|
30
|
|
|
return result.drop(["item"], axis=1) |
|
31
|
|
|
|
|
32
|
|
|
|
|
33
|
|
|
def drop_duplicates_unhashable_df(df): |
|
34
|
|
|
return df.loc[df.astype(str).drop_duplicates().index] |
|
35
|
|
|
|
|
36
|
|
|
|
|
37
|
|
|
def meta_data_path(): |
|
38
|
|
|
current_path = os.path.realpath(__file__) |
|
39
|
|
|
return current_path.rsplit("/", 1)[0] + "/" |
|
40
|
|
|
|
|
41
|
|
|
|
|
42
|
|
|
class LongTermMemory: |
|
43
|
|
|
def __init__(self, model_name, study_name=None, path=None, verbosity=None): |
|
44
|
|
|
if study_name is None: |
|
45
|
|
|
study_name = "default" |
|
46
|
|
|
|
|
47
|
|
|
model_study_name = model_name + ":" + study_name |
|
48
|
|
|
|
|
49
|
|
|
if path is None: |
|
50
|
|
|
self.ltm_data_dir = ltm_data_path() |
|
51
|
|
|
else: |
|
52
|
|
|
self.ltm_data_dir = path + "/ltm_data/" |
|
53
|
|
|
|
|
54
|
|
|
self.model_dir = self.ltm_data_dir + model_study_name + "/" |
|
55
|
|
|
|
|
56
|
|
|
if not os.path.exists(self.model_dir): |
|
57
|
|
|
os.makedirs(self.model_dir) |
|
58
|
|
|
|
|
59
|
|
|
# print("\n model_dir \n", self.model_dir) |
|
60
|
|
|
|
|
61
|
|
|
self.search_data_path = self.model_dir + "search_data.pkl" |
|
62
|
|
|
self.obj_func_path = self.model_dir + "objective_function.pkl" |
|
63
|
|
|
|
|
64
|
|
|
self.n_old_samples = 0 |
|
65
|
|
|
self.n_new_samples = 0 |
|
66
|
|
|
|
|
67
|
|
|
def remove_model_data(self): |
|
68
|
|
|
try: |
|
69
|
|
|
shutil.rmtree(self.model_dir) |
|
70
|
|
|
except OSError: |
|
71
|
|
|
pass |
|
72
|
|
|
|
|
73
|
|
|
def _dill_dump(self, object_, path): |
|
74
|
|
|
with open(path, "wb") as handle: |
|
75
|
|
|
dill.dump(object_, handle) |
|
76
|
|
|
|
|
77
|
|
|
def _dill_load(self, path): |
|
78
|
|
|
if self._pkl_valid(path): |
|
79
|
|
|
with open(path, "rb") as handle: |
|
80
|
|
|
object_ = dill.load(handle) |
|
81
|
|
|
|
|
82
|
|
|
return object_ |
|
83
|
|
|
|
|
84
|
|
|
def _pkl_valid(self, pkl_path): |
|
85
|
|
|
return os.path.isfile(pkl_path) and os.path.getsize(pkl_path) > 0 |
|
86
|
|
|
|
|
87
|
|
|
def _get_old_samples_size(self, df): |
|
88
|
|
|
if isinstance(df, pd.DataFrame): |
|
89
|
|
|
self.n_old_samples = len(df) |
|
90
|
|
|
|
|
91
|
|
|
def load_obj_func(self): |
|
92
|
|
|
return self._dill_load(self.obj_func_path) |
|
93
|
|
|
|
|
94
|
|
|
def load_search_data(self): |
|
95
|
|
|
return self._dill_load(self.search_data_path) |
|
96
|
|
|
|
|
97
|
|
|
def load(self): |
|
98
|
|
|
print("Reading in long term memory ...", end="\r") |
|
99
|
|
|
self.results_old = self._dill_load(self.search_data_path) |
|
100
|
|
|
self._get_old_samples_size(self.results_old) |
|
101
|
|
|
|
|
102
|
|
|
print( |
|
103
|
|
|
"Reading long term memory was successful:", |
|
104
|
|
|
self.n_old_samples, |
|
105
|
|
|
"samples found", |
|
106
|
|
|
) |
|
107
|
|
|
|
|
108
|
|
|
return self.results_old |
|
109
|
|
|
|
|
110
|
|
|
def save(self, dataframe, objective_function): |
|
111
|
|
|
self.results_old = self._dill_load(self.search_data_path) |
|
112
|
|
|
|
|
113
|
|
|
if self.results_old is not None: |
|
114
|
|
|
self.n_old_samples = len(self.results_old) |
|
115
|
|
|
|
|
116
|
|
|
dataframe = merge_unhashable_df(dataframe, self.results_old) |
|
117
|
|
|
dataframe = drop_duplicates_unhashable_df(dataframe).reset_index(drop=True) |
|
118
|
|
|
|
|
119
|
|
|
self.n_new_samples = len(dataframe) |
|
120
|
|
|
|
|
121
|
|
|
self._dill_dump(objective_function, self.obj_func_path) |
|
122
|
|
|
print("Saving long term memory ...", end="\r") |
|
123
|
|
|
|
|
124
|
|
|
self._dill_dump(dataframe, self.search_data_path) |
|
125
|
|
|
|
|
126
|
|
|
print( |
|
127
|
|
|
"Saving long term memory was successful:", |
|
128
|
|
|
self.n_new_samples - self.n_old_samples, |
|
129
|
|
|
"new samples found", |
|
130
|
|
|
) |
|
131
|
|
|
|
SimonBlanke /
Hyperactive
