|
1
|
|
|
#!/usr/bin/env python |
|
2
|
|
|
# coding=utf-8 |
|
3
|
|
|
from __future__ import division, print_function, unicode_literals |
|
4
|
|
|
import jsonpickle as json |
|
5
|
|
|
from jsonpickle.handlers import BaseHandler |
|
6
|
|
|
import json as _json |
|
7
|
|
|
|
|
8
|
|
|
from sacred import optional as opt |
|
9
|
|
|
|
|
10
|
|
|
__all__ = ('flatten', 'restore') |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
# class DatetimeHandler(BaseHandler): |
|
14
|
|
|
# def restore(self, obj): |
|
15
|
|
|
# return datetime.datetime.strptime(obj['date'], "%Y-%m-%dT%H:%M:%S.%f") |
|
16
|
|
|
# |
|
17
|
|
|
# def flatten(self, obj, data): |
|
18
|
|
|
# data['date'] = obj.isoformat() |
|
19
|
|
|
# return data |
|
20
|
|
|
# |
|
21
|
|
|
# DatetimeHandler.handles(datetime.datetime) |
|
22
|
|
|
|
|
23
|
|
|
if opt.has_numpy: |
|
24
|
|
|
np = opt.np |
|
25
|
|
|
|
|
26
|
|
|
class NumpyArrayHandler(BaseHandler): |
|
27
|
|
|
def flatten(self, obj, data): |
|
28
|
|
|
data['values'] = obj.tolist() |
|
29
|
|
|
data['dtype'] = str(obj.dtype) |
|
30
|
|
|
return data |
|
31
|
|
|
|
|
32
|
|
|
def restore(self, obj): |
|
33
|
|
|
return opt.np.array(obj["values"], dtype=obj["dtype"]) |
|
34
|
|
|
|
|
35
|
|
|
class NumpyGenericHandler(BaseHandler): |
|
36
|
|
|
def flatten(self, obj, data): |
|
37
|
|
|
return np.asscalar(obj) |
|
38
|
|
|
|
|
39
|
|
|
def restore(self, obj): |
|
40
|
|
|
return obj |
|
41
|
|
|
|
|
42
|
|
|
NumpyArrayHandler.handles(np.ndarray) |
|
43
|
|
|
for t in [np.bool_, np.int_, np.float_, np.intc, np.intp, np.int8, |
|
44
|
|
|
np.int16, np.int32, np.int64, np.uint8, np.uint16, np.uint32, |
|
45
|
|
|
np.uint64, np.float16, np.float32, np.float64]: |
|
46
|
|
|
NumpyGenericHandler.handles(t) |
|
47
|
|
|
|
|
48
|
|
|
|
|
49
|
|
|
if opt.has_pandas: |
|
50
|
|
|
import pandas as pd |
|
51
|
|
|
|
|
52
|
|
|
class PandasDataframeHandler(BaseHandler): |
|
53
|
|
|
def flatten(self, obj, data): |
|
54
|
|
|
# TODO: this is slow |
|
55
|
|
|
data['values'] = json.loads(obj.to_json()) |
|
56
|
|
|
data['dtypes'] = {k: str(v) for k, v in dict(obj.dtypes).items()} |
|
57
|
|
|
return data |
|
58
|
|
|
|
|
59
|
|
|
def restore(self, obj): |
|
60
|
|
|
# TODO: get rid of unnecessary json.dumps |
|
61
|
|
|
return pd.read_json(json.dumps(obj['values']), |
|
62
|
|
|
dtype=obj['dtypes']) |
|
63
|
|
|
|
|
64
|
|
|
PandasDataframeHandler.handles(pd.DataFrame) |
|
65
|
|
|
|
|
66
|
|
|
json.set_encoder_options('simplejson', sort_keys=True, indent=4) |
|
67
|
|
|
json.set_encoder_options('demjson', compactly=False) |
|
68
|
|
|
|
|
69
|
|
|
|
|
70
|
|
|
def flatten(obj): |
|
71
|
|
|
return _json.loads(json.encode(obj, keys=True)) |
|
72
|
|
|
|
|
73
|
|
|
|
|
74
|
|
|
def restore(flat): |
|
75
|
|
|
return json.decode(_json.dumps(flat), keys=True) |
|
76
|
|
|
|
IDSIA /
sacred
