141a74bba6
use COLON to avoid byte buffer error
2015-04-10 13:16:43 +08:00
e72e38bd3f
given that line_parts has been copied to features, parsing should be based on features rather than line_parts
2015-04-09 22:00:10 +08:00
249a96bc96
directly set target to empty list instead of empty string
2015-04-09 21:59:10 +08:00
8d4b9d68a3
a simple fix to reading empty label sets in the multilabel case
2015-04-09 16:53:48 +08:00
bb743ddb2a
ENH speed up SVMlight loader using Cython's array support
...
Loading the binary covtype dataset from
http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html
now takes 1/3 less wall clock time, which includes disk I/O and
decompression (bzip2).
Got rid of ArrayBuilder in the process.
2014-01-05 22:10:02 +01:00
dacfd8bd5d
DOC: Replaced all BSD style licenses with "BSD 3 clause"
...
Replaced all BSD style licenses with "BSD 3 clause"
Not checked yet!
Removed duplicate "3 clause, 3 clause"
Removed trailing period if exists
Fixed some missed licences, still about 50 to do, but those can be automated
Think I got the last of them.
Apparently me and sed have different ideas of regex.
Found a few more
2013-04-30 08:34:46 +02:00
966e0f180b
PY3: convert string literal to bytes using six in cython file
2013-03-19 18:02:39 +01:00
37c1d092f8
COSMIT shut up the build by calling np.import_array in Cython modules
...
$(make in 2>&1 | wc -l) down from 706 to 517.
Took the liberty of modernizing some of the Cython code while I was at it.
2013-01-10 18:00:21 +01:00
398fa9d136
FIX: query_id parameter and other cosmetic changes
2012-09-28 11:24:28 +02:00
6fd39fbf12
Add support for preference contraints in svmlight format.
...
This feature is explained in http://svmlight.joachims.org/ and adds
the ability to control which pairs are considered when dealing with
pairwise loss functions (e.g. for SVMrank).
In order to preserve API compatibility, I return this information (if
present) as a second column to the y vector.
2012-09-26 17:48:13 +02:00
91aaabc745
ENH micro-optimizations in SVMlight loader
...
* Some more type annotations
* Prevent (copying) string slice; instead, trust str.split
to handle trailing newlines.
2012-09-15 14:52:42 +02:00
6f54691eb3
Check that feature indices are sorted.
2012-08-05 20:34:59 +09:00
6d6c9a8937
ENH zero_based="auto" support + better n_features=None in load_svmlight_files
2012-04-10 18:20:28 +02:00
2c293ebc40
BUG re-allow zero-based indexes in SVMlight files
...
Fixes issue #750 .
2012-04-04 22:44:53 +02:00
68b7ca32cc
FIX: py3k bytes split in svmlight format parser
2011-12-26 09:32:57 +01:00
ad53ecd657
Merge branch 'master' into multilabel
...
Conflicts:
sklearn/datasets/_svmlight_format.c
Solved by re-cythonizing.
2011-12-21 11:15:33 +01:00
ef6a563f25
BUG fix off-by-one error in SVMlight format loader
...
Apparently, column indices in SVMlight/LibSVM data files are one-indexed.
2011-12-20 17:18:40 +01:00
b8085fefaa
BUG only handle labels specially in SVMlight loader + multilabel
2011-12-20 16:20:38 +01:00
c3aa2d7e8d
Merge multilabel branch with master.
...
Conflicts:
sklearn/datasets/_svmlight_format.c
sklearn/datasets/_svmlight_format.pyx
sklearn/datasets/tests/test_svmlight_format.py
sklearn/preprocessing/__init__.py
2011-12-20 12:41:42 +01:00
48fda638a5
ENH incrementally build arrays in SVMlight loader to reduce memory usage
...
Memory use on covtype.binary [1] reduced by nearly 2/3, at the expense of
of a 40% slowdown.
[1] http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html#covtype.binary
2011-11-17 19:43:00 +01:00
6f232b9309
ENH multilabel support in SVMlight loader
2011-10-27 14:05:54 +02:00
797041a86c
ENH Cython version of SVMlight loader
...
Speedup of ~30% on MNIST dataset from
http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass.html
2011-10-20 20:58:01 +02:00
Hsuan-Tien Lin