-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathCHANGELOG
95 lines (76 loc) · 3.37 KB
/
CHANGELOG
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
### v0.14.1 - 16 Dec 2022
* update TF API for newer versions
* supporting newer TF versions should solve some issues with installation via bioconda that would make it impossible to install the latest version under some configurations when using tensorflow-gpu
* warn when preprocessing sequences longer than expected and ask the user to explicitly enable trimming (--trim)
* add negative and undefined outputs in filter: now you can automatically get also the non-positive reads (see rki_bioinformatics/DeePaC#3).
* small refactors and bugfixes
### v0.14.0 - 18 Nov 2021
* new feature: multiclass models
### v0.13.6 - 03 Feb 2021
* bugfix: tf.dataset preprocessing
### v0.13.5 - 24 Jan 2021
* setup bugfix
### v0.13.4 - 23 Jan 2021
* fetching new models from the deepac model repo
### v0.13.3 - 25 Nov 2020
* bugfix in eval
### v0.13.2 - 19 Nov 2020
* Reorder imports for bioconda mulled test compatibility
### v0.13.1 - 18 Nov 2020
* bugfix: dropout would be disabled if MC dropout was
### v0.13.0 - 13 Oct 2020
* skip connections
* confidence thresholds for predictions
* MC dropout for Bayesian uncertainty
* custom weight initialization schemes
* bugfix: RC-constraint for "Siam" nets
* bugfix: Fwd and rc filter instances in filter contributions
* bugfix: correct custom weight initialization
* minor bugfixes
### v0.12.2 - 13 Jul 2020
* utilities: gff3 to .genome conversion tool, get builtin models
* extended filter enrichment analyses
* L1 option
* bugfixes
* UI/config improvements (last epoch is inclusive)
### v0.12.1 - 2 Jul 2020
* Treat forward and reverse-complement instances of filters separately when checking filter enrichment (detects more instances)
* Bugfixes
### v0.12.0 - 15 May 2020
* Extended multi-GPU support, easy GPU placement
* TPU support (experimental)
* Alternative distributed computation strategies (experimental)
* TFRecord training
* Cleaner config files, CLI improvements
* Additional debugging options
* Bugfixes
### v0.11.0 - 5 May 2020
* Full TensorFlow 2 compatibility (eager exec disabled for SHAP)
* Generate templates with deepac templates
* CLI tweaks
* Dynamic subread generation in training
* Bugfixes
### v0.10.1 - 20 Feb 2020
* Added workaround for a libgomp importing bug present on some systems (including bioconda's mulled-test containers)
### v0.10.0 - 19 Feb 2020
New functionalities:
* convolutional filter visualization with partial Shapley values (max-contribution or all-contributions) or max-activation
* genome-wide phenotype potential maps
* nucleotide contribution maps
* filter enrichment analysis
* filter and gene rankings
Other:
* TF 1.15 required; TF 2.0 compatible with training and prediction but untested, as TFExplainer in SHAP not yet compatible (so the intepretability workflows require TF<2.0)
* Minor bugfixes
### v0.9.3 - 13 May 2019
* Automated tests
* Bugfix: always saving models trained on built-in architectures
### v0.9.2 - 15 Apr 2019
* Windows compatibility
* CLI bugfixes
* Python3 required
* Built-in architectures for training
### v0.9.1 - 12 Apr 2019
* added the CPU version of TensorFlow as the default requirement
### v0.9.0 - 3 Apr 2019
* Initial PyPi release