1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
|
"""Automatically download MLdata datasets."""
# Copyright (c) 2011 Pietro Berkes
# License: Simplified BSD
import os
from os.path import join, exists
import re
import scipy as sp
from scipy import io
from shutil import copyfileobj
import urllib2
from .base import get_data_home, Bunch
MLDATA_BASE_URL = "http://mldata.org/repository/data/download/matlab/%s"
def mldata_filename(dataname):
"""Convert a raw name for a data set in a mldata.org filename."""
dataname = dataname.lower().replace(' ', '-')
return re.sub(r'[().]', '', dataname)
def fetch_mldata(dataname, target_name='label', data_name='data',
transpose_data=True, data_home=None):
"""Fetch an mldata.org data set
If the file does not exist yet, it is downloaded from mldata.org .
mldata.org does not have an enforced convention for storing data or
naming the columns in a data set. The default behavior of this function
works well with the most common cases:
1) data values are stored in the column 'data', and target values in the
column 'label'
2) alternatively, the first column stores target values, and the second
data values
3) the data array is stored as `n_features x n_samples` , and thus needs
to be transposed to match the `sklearn` standard
Keyword arguments allow to adapt these defaults to specific data sets
(see parameters `target_name`, `data_name`, `transpose_data`, and
the examples below).
mldata.org data sets may have multiple columns, which are stored in the
Bunch object with their original name.
Parameters
----------
dataname:
Name of the data set on mldata.org,
e.g.: "leukemia", "Whistler Daily Snowfall", etc.
The raw name is automatically converted to a mldata.org URL .
target_name: optional, default: 'label'
Name or index of the column containing the target values.
data_name: optional, default: 'data'
Name or index of the column containing the data.
transpose_data: optional, default: True
If True, transpose the downloaded data array.
data_home: optional, default: None
Specify another download and cache folder for the data sets. By default
all scikit learn data is stored in '~/scikit_learn_data' subfolders.
Returns
-------
data : Bunch
Dictionary-like object, the interesting attributes are:
'data', the data to learn, 'target', the classification labels,
'DESCR', the full description of the dataset, and
'COL_NAMES', the original names of the dataset columns.
Examples
--------
Load the 'iris' dataset from mldata.org:
>>> from sklearn.datasets.mldata import fetch_mldata
>>> iris = fetch_mldata('iris')
>>> iris.target[0]
1
>>> print iris.data[0]
[-0.555556 0.25 -0.864407 -0.916667]
Load the 'leukemia' dataset from mldata.org, which respects the
sklearn axes convention:
>>> leuk = fetch_mldata('leukemia', transpose_data=False)
>>> print leuk.data.shape[0]
7129
Load an alternative 'iris' dataset, which has different names for the
columns:
>>> iris2 = fetch_mldata('datasets-UCI iris', target_name=1,
... data_name=0)
>>> iris3 = fetch_mldata('datasets-UCI iris',
... target_name='class', data_name='double0')
"""
# normalize dataset name
dataname = mldata_filename(dataname)
# check if this data set has been already downloaded
data_home = get_data_home(data_home=data_home)
data_home = join(data_home, 'mldata')
if not exists(data_home):
os.makedirs(data_home)
matlab_name = dataname + '.mat'
filename = join(data_home, matlab_name)
# if the file does not exist, download it
if not exists(filename):
urlname = MLDATA_BASE_URL % urllib2.quote(dataname)
try:
mldata_url = urllib2.urlopen(urlname)
except urllib2.HTTPError as e:
if e.code == 404:
e.msg = "Dataset '%s' not found on mldata.org." % dataname
raise
# store Matlab file
try:
with open(filename, 'w+b') as matlab_file:
copyfileobj(mldata_url, matlab_file)
except:
os.remove(filename)
raise
mldata_url.close()
# load dataset matlab file
with open(filename, 'rb') as matlab_file:
matlab_dict = io.loadmat(matlab_file, struct_as_record=True)
# -- extract data from matlab_dict
# flatten column names
col_names = [str(descr[0])
for descr in matlab_dict['mldata_descr_ordering'][0]]
# if target or data names are indices, transform then into names
if isinstance(target_name, int):
target_name = col_names[target_name]
if isinstance(data_name, int):
data_name = col_names[data_name]
# rules for making sense of the mldata.org data format
# (earlier ones have priority):
# 1) there is only one array => it is "data"
# 2) there are multiple arrays
# a) copy all columns in the bunch, using their column name
# b) if there is a column called `target_name`, set "target" to it,
# otherwise set "target" to first column
# c) if there is a column called `data_name`, set "data" to it,
# otherwise set "data" to second column
dataset = {'DESCR': 'mldata.org dataset: %s' % dataname,
'COL_NAMES': col_names}
# 1) there is only one array => it is considered data
if len(col_names) == 1:
data_name = col_names[0]
dataset['data'] = matlab_dict[data_name]
# 2) there are multiple arrays
else:
for name in col_names:
dataset[name] = matlab_dict[name]
if target_name in col_names:
del dataset[target_name]
dataset['target'] = matlab_dict[target_name]
else:
del dataset[col_names[0]]
dataset['target'] = matlab_dict[col_names[0]]
if data_name in col_names:
del dataset[data_name]
dataset['data'] = matlab_dict[data_name]
else:
del dataset[col_names[1]]
dataset['data'] = matlab_dict[col_names[1]]
# set axes to sklearn conventions
if transpose_data:
dataset['data'] = dataset['data'].T
if 'target' in dataset:
if not sp.sparse.issparse(dataset['target']):
dataset['target'] = dataset['target'].squeeze()
return Bunch(**dataset)
|
