_rcv1.py
10.6 KB
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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
"""RCV1 dataset.
The dataset page is available at
http://jmlr.csail.mit.edu/papers/volume5/lewis04a/
"""
# Author: Tom Dupre la Tour
# License: BSD 3 clause
import logging
from os import remove, makedirs
from os.path import dirname, exists, join
from gzip import GzipFile
import numpy as np
import scipy.sparse as sp
import joblib
from . import get_data_home
from ._base import _pkl_filepath
from ._base import _fetch_remote
from ._base import RemoteFileMetadata
from ._svmlight_format_io import load_svmlight_files
from ..utils import shuffle as shuffle_
from ..utils import Bunch
from ..utils.validation import _deprecate_positional_args
# The original vectorized data can be found at:
# http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/a13-vector-files/lyrl2004_vectors_test_pt0.dat.gz
# http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/a13-vector-files/lyrl2004_vectors_test_pt1.dat.gz
# http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/a13-vector-files/lyrl2004_vectors_test_pt2.dat.gz
# http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/a13-vector-files/lyrl2004_vectors_test_pt3.dat.gz
# http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/a13-vector-files/lyrl2004_vectors_train.dat.gz
# while the original stemmed token files can be found
# in the README, section B.12.i.:
# http://www.ai.mit.edu/projects/jmlr/papers/volume5/lewis04a/lyrl2004_rcv1v2_README.htm
XY_METADATA = (
RemoteFileMetadata(
url='https://ndownloader.figshare.com/files/5976069',
checksum=('ed40f7e418d10484091b059703eeb95a'
'e3199fe042891dcec4be6696b9968374'),
filename='lyrl2004_vectors_test_pt0.dat.gz'),
RemoteFileMetadata(
url='https://ndownloader.figshare.com/files/5976066',
checksum=('87700668ae45d45d5ca1ef6ae9bd81ab'
'0f5ec88cc95dcef9ae7838f727a13aa6'),
filename='lyrl2004_vectors_test_pt1.dat.gz'),
RemoteFileMetadata(
url='https://ndownloader.figshare.com/files/5976063',
checksum=('48143ac703cbe33299f7ae9f4995db4'
'9a258690f60e5debbff8995c34841c7f5'),
filename='lyrl2004_vectors_test_pt2.dat.gz'),
RemoteFileMetadata(
url='https://ndownloader.figshare.com/files/5976060',
checksum=('dfcb0d658311481523c6e6ca0c3f5a3'
'e1d3d12cde5d7a8ce629a9006ec7dbb39'),
filename='lyrl2004_vectors_test_pt3.dat.gz'),
RemoteFileMetadata(
url='https://ndownloader.figshare.com/files/5976057',
checksum=('5468f656d0ba7a83afc7ad44841cf9a5'
'3048a5c083eedc005dcdb5cc768924ae'),
filename='lyrl2004_vectors_train.dat.gz')
)
# The original data can be found at:
# http://jmlr.csail.mit.edu/papers/volume5/lewis04a/a08-topic-qrels/rcv1-v2.topics.qrels.gz
TOPICS_METADATA = RemoteFileMetadata(
url='https://ndownloader.figshare.com/files/5976048',
checksum=('2a98e5e5d8b770bded93afc8930d882'
'99474317fe14181aee1466cc754d0d1c1'),
filename='rcv1v2.topics.qrels.gz')
logger = logging.getLogger(__name__)
@_deprecate_positional_args
def fetch_rcv1(*, data_home=None, subset='all', download_if_missing=True,
random_state=None, shuffle=False, return_X_y=False):
"""Load the RCV1 multilabel dataset (classification).
Download it if necessary.
Version: RCV1-v2, vectors, full sets, topics multilabels.
================= =====================
Classes 103
Samples total 804414
Dimensionality 47236
Features real, between 0 and 1
================= =====================
Read more in the :ref:`User Guide <rcv1_dataset>`.
.. versionadded:: 0.17
Parameters
----------
data_home : string, optional
Specify another download and cache folder for the datasets. By default
all scikit-learn data is stored in '~/scikit_learn_data' subfolders.
subset : string, 'train', 'test', or 'all', default='all'
Select the dataset to load: 'train' for the training set
(23149 samples), 'test' for the test set (781265 samples),
'all' for both, with the training samples first if shuffle is False.
This follows the official LYRL2004 chronological split.
download_if_missing : boolean, default=True
If False, raise a IOError if the data is not locally available
instead of trying to download the data from the source site.
random_state : int, RandomState instance, default=None
Determines random number generation for dataset shuffling. Pass an int
for reproducible output across multiple function calls.
See :term:`Glossary <random_state>`.
shuffle : bool, default=False
Whether to shuffle dataset.
return_X_y : boolean, default=False.
If True, returns ``(dataset.data, dataset.target)`` instead of a Bunch
object. See below for more information about the `dataset.data` and
`dataset.target` object.
.. versionadded:: 0.20
Returns
-------
dataset : :class:`~sklearn.utils.Bunch`
Dictionary-like object, with the following attributes.
data : scipy csr array, dtype np.float64, shape (804414, 47236)
The array has 0.16% of non zero values.
target : scipy csr array, dtype np.uint8, shape (804414, 103)
Each sample has a value of 1 in its categories, and 0 in others.
The array has 3.15% of non zero values.
sample_id : numpy array, dtype np.uint32, shape (804414,)
Identification number of each sample, as ordered in dataset.data.
target_names : numpy array, dtype object, length (103)
Names of each target (RCV1 topics), as ordered in dataset.target.
DESCR : string
Description of the RCV1 dataset.
(data, target) : tuple if ``return_X_y`` is True
.. versionadded:: 0.20
"""
N_SAMPLES = 804414
N_FEATURES = 47236
N_CATEGORIES = 103
N_TRAIN = 23149
data_home = get_data_home(data_home=data_home)
rcv1_dir = join(data_home, "RCV1")
if download_if_missing:
if not exists(rcv1_dir):
makedirs(rcv1_dir)
samples_path = _pkl_filepath(rcv1_dir, "samples.pkl")
sample_id_path = _pkl_filepath(rcv1_dir, "sample_id.pkl")
sample_topics_path = _pkl_filepath(rcv1_dir, "sample_topics.pkl")
topics_path = _pkl_filepath(rcv1_dir, "topics_names.pkl")
# load data (X) and sample_id
if download_if_missing and (not exists(samples_path) or
not exists(sample_id_path)):
files = []
for each in XY_METADATA:
logger.info("Downloading %s" % each.url)
file_path = _fetch_remote(each, dirname=rcv1_dir)
files.append(GzipFile(filename=file_path))
Xy = load_svmlight_files(files, n_features=N_FEATURES)
# Training data is before testing data
X = sp.vstack([Xy[8], Xy[0], Xy[2], Xy[4], Xy[6]]).tocsr()
sample_id = np.hstack((Xy[9], Xy[1], Xy[3], Xy[5], Xy[7]))
sample_id = sample_id.astype(np.uint32, copy=False)
joblib.dump(X, samples_path, compress=9)
joblib.dump(sample_id, sample_id_path, compress=9)
# delete archives
for f in files:
f.close()
remove(f.name)
else:
X = joblib.load(samples_path)
sample_id = joblib.load(sample_id_path)
# load target (y), categories, and sample_id_bis
if download_if_missing and (not exists(sample_topics_path) or
not exists(topics_path)):
logger.info("Downloading %s" % TOPICS_METADATA.url)
topics_archive_path = _fetch_remote(TOPICS_METADATA,
dirname=rcv1_dir)
# parse the target file
n_cat = -1
n_doc = -1
doc_previous = -1
y = np.zeros((N_SAMPLES, N_CATEGORIES), dtype=np.uint8)
sample_id_bis = np.zeros(N_SAMPLES, dtype=np.int32)
category_names = {}
with GzipFile(filename=topics_archive_path, mode='rb') as f:
for line in f:
line_components = line.decode("ascii").split(" ")
if len(line_components) == 3:
cat, doc, _ = line_components
if cat not in category_names:
n_cat += 1
category_names[cat] = n_cat
doc = int(doc)
if doc != doc_previous:
doc_previous = doc
n_doc += 1
sample_id_bis[n_doc] = doc
y[n_doc, category_names[cat]] = 1
# delete archive
remove(topics_archive_path)
# Samples in X are ordered with sample_id,
# whereas in y, they are ordered with sample_id_bis.
permutation = _find_permutation(sample_id_bis, sample_id)
y = y[permutation, :]
# save category names in a list, with same order than y
categories = np.empty(N_CATEGORIES, dtype=object)
for k in category_names.keys():
categories[category_names[k]] = k
# reorder categories in lexicographic order
order = np.argsort(categories)
categories = categories[order]
y = sp.csr_matrix(y[:, order])
joblib.dump(y, sample_topics_path, compress=9)
joblib.dump(categories, topics_path, compress=9)
else:
y = joblib.load(sample_topics_path)
categories = joblib.load(topics_path)
if subset == 'all':
pass
elif subset == 'train':
X = X[:N_TRAIN, :]
y = y[:N_TRAIN, :]
sample_id = sample_id[:N_TRAIN]
elif subset == 'test':
X = X[N_TRAIN:, :]
y = y[N_TRAIN:, :]
sample_id = sample_id[N_TRAIN:]
else:
raise ValueError("Unknown subset parameter. Got '%s' instead of one"
" of ('all', 'train', test')" % subset)
if shuffle:
X, y, sample_id = shuffle_(X, y, sample_id, random_state=random_state)
module_path = dirname(__file__)
with open(join(module_path, 'descr', 'rcv1.rst')) as rst_file:
fdescr = rst_file.read()
if return_X_y:
return X, y
return Bunch(data=X, target=y, sample_id=sample_id,
target_names=categories, DESCR=fdescr)
def _inverse_permutation(p):
"""inverse permutation p"""
n = p.size
s = np.zeros(n, dtype=np.int32)
i = np.arange(n, dtype=np.int32)
np.put(s, p, i) # s[p] = i
return s
def _find_permutation(a, b):
"""find the permutation from a to b"""
t = np.argsort(a)
u = np.argsort(b)
u_ = _inverse_permutation(u)
return t[u_]