-
Notifications
You must be signed in to change notification settings - Fork 19
/
CHANGES.txt
176 lines (140 loc) · 7.8 KB
/
CHANGES.txt
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
************************************************
DKPro-TC 1.0.0 Release
== Highlights ==
#460 - Add XGBoost as MLA
#429 - Replace evaluation module
#388 - Add support for deep learning frameworks
== Major Version Enhancements ==
DKPro Core 1.9.1
Liblinear 2.20
DKPro Lab 0.14.0
== Other Issues Fixed ==
#477 - Weka: Separate MLA code from DKPro Lab
#476 - SvmHmm: Separate MLA code from DKPro Lab
#473 - Xgboost: Separate MLA code from DKPro Lab
#474 - Libsvm: Separate MLA code from DKPro Lab
#471 - Liblinear: Separate MLA code from DKPro Lab
#470 - CrfSuite: Separate MLA code from DKPro Lab
************************************************
DKPro-TC 0.9.0 Release
== Highlights ==
Issue 358: Speed up sequence tagging by using sparse-feature mode as default
Issue 355: Make document mode to a special case of unit mode
Issue 353: Enable passing CollectionReaderDescriptions instead of defining reader dimensions
Issue 341: Improve DKPro-TC preprocessing speed by removing CasMultiplier from Workflow
Issue 293: Add mavenized libsvm as machine learning adapter
Issue 50: Create a dedicated evaluation object/module
Issue 39: Multiple instances of the same feature extractor with different configurations should be possible
== Major Version Enhancements ==
DKPro Lab 0.13.0
DKPro Core 1.8.0
Meka 1.9.0
== Other Issues Fixed ==
Issue 363: Extend basic-implementation of Liblinear
Issue 367: CRFsuite: Enable full parametrization
Issue 366: Unify parametrization of SVM modules
Issue 361: Refactor ValidityCheckConnector (Pre/Post)
Issue 207: Potential severe problem with ngram meta collectors and extractors
************************************************
DKPro-TC 0.8.0 Release
== Highlights ==
Issue 223: Add support for model load/save on Weka/CRFsuite
Issue 282: Model saving/loading of SVMHMM models
Issue 296: Adding save-model TaskBase to TCMachineLearningAdapter
== Major Version Enhancements ==
DKPro Lab 0.12.0
DKPro parent pom 14
Java 8
== Other Issues Fixed ==
Issue 239: SVMhmm is missing OSx binary
Issue 269: Replace @author annotation in source by a contributors.txt file
Issue 295: Setting NoOpAnnotator by default if no preprocessing is defined
Issue 305: CRFSuite binary: Windows 64 bit binary is missing
Issue 321: Update package names to org.dkpro.tc
Issue 326: Example Project: Prefixing examples with the MLA name that is used
************************************************
DKPro-TC 0.7.0 Release
== Highlights ==
Issue 183: support for CRFsuite machine learning framework
Issue 190: Support for SVMhmm machine learning framework
Issue 126: new generic machine learning module dkpro-tc-ml, new functionality to set a machine learning adapter
Issues 199, 200: New functionality to set the feature store, new sparse feature store
== Enhancements ==
Additional demos for new machine learning frameworks and other experiment setups
New module dkpro-tc-evaluation for cross-framework evaluation functionality (not fully supported yet, only works for single-label learning mode)
== Major Version Enhancements ==
Upgrade to DKPro Core 1.7.0 / DKPro parent pom 10
Upgrade to Weka 3.7.11
Upgrade to Meka 1.7.3
== Other Issues Fixed ==
Issue 220: Fixed wrong calculation of label-wise scores in multi-label learning mode
Issue 163: SimpleDkproTCReader does not allow to set language
Issue 139: Experiments should fail with a meaningful exception if no data are found
Issue 160: Invalid column index in reports which produce XLS files
Issue 175: ValidityCheckTask only verifies the experiment setup based on training data
== Other ==
Module de.tudarmstadt.ukp.dkpro.tc.features.pair.similarity-asl has been excluded from the Maven Central release, since the "DKPro Similarity" dependency is not yet available on Central.
All demos in dkpro-tc-examples and dkpro-tc-examples-groovy now have unit tests.
dkpro-tc-ml-mallet is deprecated and will be removed from the next release due to a lack of support.
************************************************
DKPro-TC 0.6.0 Release
== Highlights ==
Release on Maven Central
Issue 107: Support sequence labeling (with Mallet)
Issue 97: Enable Modes for Learning and Feature Types
NGram and PairNGram modules have been entirely re-written
Demo modules have been re-ordered: separated modules for Groovy and Java setup
Upgrade to Java 7
== Enhancements ==
Support for SkipNGrams
Issue 117: Coverage of source code comments has been increased
Issue 79: Support for Missing Values in Feature Extractors
Issue 24: Bag of sounds feature extractor
Issue 90: Add data writers for testing
Issue 104: Add PerformanceBatchReport
Increased number of demos (e.g. for unit classification )
Added BatchTaskCrossValidationWithFoldControl
== Major Version Enhancements ==
Upgrade to DKPro Lab 0.11.0
Upgrade to DKPro Core 1.6.1
Upgrade to Apache uimafit 2.1.0
Upgrade to Meka 1.6.2
Upgrade to Groovy 2.2.2
== Other Issues Fixed ==
Issue 80: Feature extractors returning empty lists can cause errors
Issue 84: DocumentFeatureExtractor and ClassificationUnitFeatureExtractor are used inconsistently
Issue 94: Setting number of folds accidentally to 1 causes exception
Issue 105: FeatureValuesReport broken
Issue 106: Throw Exception when mode-incompatible FE is encountered
Issue 116 Improve format of the output by OutcomeIDReport
Issue 118: Rename preprocessing aggregate
Issue 131: NrOfTokensFEs cause an exception in FeatureExtractionTask when at least one but not all documents contain valid sentence annotations
Issue 133: Phonetic ngrams in POS ngram DFE can't be turned off? Crash on some characters
Issue 142: Some lucene ngram Meta Collectors don't record term counts in a document
Issue 152: Naming scheme of ArtifactIds changed to the simpler, dashed Maven standard. Furtermore, license suffixes have been removed from ArtifactIDs; e.g. "dkpro-tc-core" instead of "de.tudarmstadt.ukp.dkpro.tc.core-asl". GroupIDs and class names remain unchanged.
Issue 156: Adding license headers to all classes
== Other ==
Module de.tudarmstadt.ukp.dkpro.tc.features.pair.similarity-asl has been excluded from the Maven Central release, since the "DKPro Similarity" dependency is not yet available on Central.
************************************************
DKPro-TC 0.5.0 Release
== Highlights ==
Re-Implementation of crossvalidation setup to do a better job
Major reduction of the memory footprint during feature extraction (issue 75)
Several new modules (de.tudarmstadt.ukp.dkpro.tc.features.ngram-asl, de.tudarmstadt.ukp.dkpro.tc.fstore.simple-asl, de.tudarmstadt.ukp.dkpro.tc.demo.pairtwentynewsgroups-gpl)
== Enhancements ==
Feature Selection with Weka is now possible (issue 49)
A show case for pair classification has been added (issue 61), along with several improvements and bugfixes for the pair mode (issue 76, 81)
A show case for running DKPro-TC experiments on a server has been added (issue 52), see http://code.google.com/p/dkpro-tc/wiki/ServerSetup
A skip ngram feature extractor has been added (issue 74)
Added Lucene-based ngram extractor (related issues: 85)
Added several parameters to make behaviour of ngram stopword filtering configurable (issues 82, 86, 87)
== Fixed Issues ==
Issue 27 (CrossValidation setup is now properly separating training and test data)
Issue 30 (ConfigurationParameters are now used according to best practice)
Issue 66 (BatchTasks are more verbose)
Issue 71 (TrainTestReport was renamed to ClassificationReport)
Issue 67 (Evaluation measure names in reports are now constants)
Issue 64 and 65 (Fixed problems in the instantiation/readers of Meta- and ExtractFeaturesTask)
Issue 54: Fixed regular expression resulting in document multiplication in the MetaTask (crossvalidation setup)
Issue 70: Fixed DocumentID setup in the Regression demo (70)
Several minor issues (53, 78, 72)