|
1
|
|
|
#! /usr/bin/env python |
|
|
|
|
|
|
2
|
|
|
# |
|
3
|
|
|
# Copyright (C) 2016 Rich Lewis <[email protected]> |
|
4
|
|
|
# License: 3-clause BSD |
|
5
|
|
|
|
|
6
|
1 |
|
import warnings |
|
7
|
1 |
|
import tempfile |
|
8
|
1 |
|
import os |
|
9
|
|
|
|
|
10
|
1 |
|
import pandas as pd |
|
|
|
|
|
|
11
|
1 |
|
import h5py |
|
|
|
|
|
|
12
|
1 |
|
from fuel.datasets import H5PYDataset |
|
|
|
|
|
|
13
|
1 |
|
from fuel.utils import find_in_data_path |
|
|
|
|
|
|
14
|
1 |
|
from fuel import config |
|
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
|
|
17
|
1 |
|
class Dataset(H5PYDataset): |
|
18
|
|
|
|
|
19
|
|
|
""" Abstract base class providing an interface to the skchem data format.""" |
|
20
|
|
|
|
|
21
|
1 |
|
def __init__(self, **kwargs): |
|
|
|
|
|
|
22
|
|
|
kwargs.setdefault('load_in_memory', True) |
|
23
|
|
|
super(Dataset, self).__init__( |
|
24
|
|
|
file_or_path=find_in_data_path(self.filename), **kwargs) |
|
|
|
|
|
|
25
|
|
|
|
|
26
|
1 |
|
@classmethod |
|
27
|
|
|
def available_sources(cls): |
|
|
|
|
|
|
28
|
|
|
with h5py.File(find_in_data_path(cls.filename)) as f: |
|
|
|
|
|
|
29
|
|
|
return cls.get_all_sources(f) |
|
|
|
|
|
|
30
|
|
|
|
|
31
|
1 |
|
@classmethod |
|
32
|
|
|
def available_sets(cls): |
|
|
|
|
|
|
33
|
|
|
with h5py.File(find_in_data_path(cls.filename)) as f: |
|
|
|
|
|
|
34
|
|
|
return cls.get_all_splits(f) |
|
|
|
|
|
|
35
|
|
|
|
|
36
|
1 |
|
@classmethod |
|
37
|
1 |
|
def load_set(cls, set_name, sources=()): |
|
38
|
|
|
|
|
39
|
|
|
""" Load the sources for a single set. |
|
40
|
|
|
|
|
41
|
|
|
Args: |
|
42
|
|
|
set_name (str): |
|
43
|
|
|
The set name. |
|
44
|
|
|
sources (tuple[str]): |
|
45
|
|
|
The sources to return data for. |
|
46
|
|
|
|
|
47
|
|
|
Returns: |
|
48
|
|
|
tuple[np.array] |
|
49
|
|
|
The requested sources for the requested set. |
|
50
|
|
|
""" |
|
51
|
|
|
if set_name == 'all': |
|
52
|
|
|
set_name = cls.set_names |
|
|
|
|
|
|
53
|
|
|
else: |
|
54
|
|
|
set_name = (set_name,) |
|
55
|
|
|
if sources == 'all': |
|
56
|
|
|
sources = cls.sources_names |
|
|
|
|
|
|
57
|
|
|
return cls(which_sets=set_name, sources=sources, load_in_memory=True).data_sources |
|
|
|
|
|
|
58
|
|
|
|
|
59
|
1 |
|
@classmethod |
|
60
|
1 |
|
def load_data(cls, sets=(), sources=()): |
|
61
|
|
|
|
|
62
|
|
|
""" Load a set of sources. |
|
63
|
|
|
|
|
64
|
|
|
Args: |
|
65
|
|
|
sets (tuple[str]): |
|
66
|
|
|
The sets to return data for. |
|
67
|
|
|
sources: |
|
68
|
|
|
The sources to return data for. |
|
69
|
|
|
|
|
70
|
|
|
Example: |
|
71
|
|
|
(X_train, y_train), (X_test, y_test) = Dataset.load_data(sets=('train', 'test'), sources=('X', 'y')) |
|
|
|
|
|
|
72
|
|
|
""" |
|
73
|
|
|
|
|
74
|
|
|
for set_name in sets: |
|
75
|
|
|
yield cls.load_set(set_name, sources) |
|
76
|
|
|
|
|
77
|
1 |
|
@classmethod |
|
78
|
|
|
def read_frame(cls, key, *args, **kwargs): |
|
79
|
|
|
|
|
80
|
|
|
""" Load a set of features from the dataset as a pandas object. |
|
81
|
|
|
|
|
82
|
|
|
Args: |
|
83
|
|
|
key (str): |
|
84
|
|
|
The HDF5 key for required data. Typically, this will be one of |
|
85
|
|
|
|
|
86
|
|
|
- structure: for the raw molecules |
|
87
|
|
|
- smiles: for the smiles |
|
88
|
|
|
- features/{feat_name}: for the features |
|
89
|
|
|
- targets/{targ_name}: for the targets |
|
90
|
|
|
|
|
91
|
|
|
Returns: |
|
92
|
|
|
pd.Series or pd.DataFrame or pd.Panel |
|
93
|
|
|
The data as a dataframe. |
|
94
|
|
|
""" |
|
95
|
|
|
|
|
96
|
1 |
|
with warnings.catch_warnings(): |
|
97
|
1 |
|
warnings.simplefilter('ignore') |
|
98
|
1 |
|
data = pd.read_hdf(find_in_data_path(cls.filename), key, *args, **kwargs) |
|
|
|
|
|
|
99
|
1 |
|
if isinstance(data, pd.Panel): |
|
100
|
|
|
data = data.transpose(2, 1, 0) |
|
101
|
1 |
|
return data |
|
102
|
|
|
|
|
103
|
1 |
|
@classmethod |
|
104
|
1 |
|
def download(cls, output_directory=None, download_directory=None): |
|
105
|
|
|
|
|
106
|
|
|
""" Download the dataset and convert it. |
|
107
|
|
|
|
|
108
|
|
|
Args: |
|
109
|
|
|
output_directory (str): |
|
110
|
|
|
The directory to save the data to. Defaults to the first |
|
111
|
|
|
directory in the fuel data path. |
|
112
|
|
|
|
|
113
|
|
|
download_directory (str): |
|
114
|
|
|
The directory to save the raw files to. Defaults to a temporary |
|
115
|
|
|
directory. |
|
116
|
|
|
|
|
117
|
|
|
Returns: |
|
118
|
|
|
str: |
|
119
|
|
|
The path of the downloaded and processed dataset. |
|
120
|
|
|
""" |
|
121
|
|
|
|
|
122
|
|
|
if not output_directory: |
|
123
|
|
|
output_directory = config.config['data_path']['yaml'].split(':')[0] |
|
124
|
|
|
|
|
125
|
|
|
output_directory = os.path.expanduser(output_directory) |
|
126
|
|
|
|
|
127
|
|
|
if not download_directory: |
|
128
|
|
|
download_directory = tempfile.mkdtemp() |
|
129
|
|
|
|
|
130
|
|
|
cls.downloader.download(directory=download_directory) |
|
|
|
|
|
|
131
|
|
|
return cls.converter.convert(directory=download_directory, |
|
|
|
|
|
|
132
|
|
|
output_directory=output_directory) |
|
133
|
|
|
|
richlewis42 /
scikit-chem
Loading history...
The coding style of this project requires that you add a docstring to this code element. Below, you find an example for methods:
If you would like to know more about docstrings, we recommend to read PEP-257: Docstring Conventions.