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
|
Changes since 1.00
2022/04/08: 1.10-1
. Added citations to DESCRIPTION and to help file for randomGLM
2021/06/12: 1.10
. Bump version for release
2021/05/30: 1.09-992
. Minor changes in internal code handling multi-threading (via cluster
functions from parallel) on Windows.
. Cleanup of metadata (DESCRIPTION) and help files to comply with CRAN
rules
. New argument 'weight' lets the user specify sample weights for the
underlying regression models.
. DIFF: randomGLM now works with categorical data. Categorical variables are
first binarized, then used as other numeric variables except their
power in interactions can never be more than 1. This may lead to
slightly different predictions when some features are categorical.
. New arguments categoricalColumns and maxCategoricalLevels allow
the user to specify which columns should be treated as categorical.
. New argument includeSelfinteractions allows the user to turn off
self-interactions for continuous features. Older versions of
randomGLM allowed self-interactions but other statistical approaches
usually exclude them.
. New argument replaceBadBagFeatures optionally turns on a more
efficient handling of features with missing data. The old behaviour
was to remove them from each bag without attempting to replace them;
the new, optional behaviour is to only sample from features that
have no missing values in the bag samples. This should lead to more
in-bag variables when missing data are present.
. Bug fix: randomGLM now correctly records variables that were
retained by the model selection when missing data are present. This
bug led to incorrect results from functions predict and
thinRandomGLM.
. Function accuracyMeasures now also works with factors.
2013/05/22: 1.05-2
. Bugfix: function works correctly with suvival objects.
. Internal bugfix that may lead to slightly different sets of features
being selected for each bag.
2013/05/09: 1.05-1
. More descriptive error message generated if all features are removed
from a bag due to missing data or constantness.
. Bug fix: predictor now works with a test set that consists of a
single sample.
2013/02/18: 1.05
. The predictor now works with a variety of responses including all
glm families and censored time (Surv) responses.
2013/02/06: 1.03
. New function accuracyMeasures for convenient calculation of most
accuracy measures.
2013/01/28: 1.02-1
. Fixes to problems with feature names that are not valid R names.
. Fixed a problem when using interactions with large numbers of
features. This required a switch from the gtools' function
combinations to an internal version; hence gtools is not required
anymore.
2013/01/23: 1.02
. Coefficient names in returned models and in
featuresInForwardRegression are now easier to read for humans. New
argument interactionSeparatorForCoefNames lets the user specify how
interaction names are formed.
2013/01/20: 1.01
. New argument minInBagObs lets user specify the minimum number of
unique in-bag observations for each bag. This prevents bags with too
few observations that can cause problems when fitting models.
. Function now works even when there is a single OOB observation in
some bags
2013/01/15: 1.00-2
. Fixed a failure of multi-threaded calculation with more than 100 bags.
2013/01/13: 1.00-1
. Fixed a bug with default (NULL) nThreads.
Changes to version included in WGCNA
. Parallelized main loop; number of threads can be controled via
argument nThreads
. bagObsIndex element of output is now nObservationsPerBag times nBags
(transposed to what it was before)
|
