Releases: Oxid15/cascade
Releases · Oxid15/cascade
v0.10.0 - Update
New Cascade version is here! New metadata viewing tools are now available, tools for logging object states and more
Adds
DataRegistrator
- tool for logging history of data #148HistoryLogger
- tool for logging history of meta of any objectsDiffViewer
- tool to view raw metadata from repos, data histories and version logs in one place #149 #151- The ability to choose metric in
HistoryViewer
s GUI #150
Refactors
- More thorough typing, especially with meta
Fixes and changes
- Important - version log format in
VersionAssigner
changed and old logs are now not supported - Important - plotly is now an optional dependency
PR's
- Update README by @Oxid15 in #140
- Documentation update by @Oxid15 in #141
- Optional plotly by @Oxid15 in #142
- Add .yaml as format that also supported since it is another name of YAML by @Oxid15 in #143
- Meta typing by @Oxid15 in #144
- ModelRepo logging by @Oxid15 in #145
- Data registration by @Oxid15 in #146
- Fix weighed sampler by @Oxid15 in #147
- Dataset history by @Oxid15 in #148
- View history by @Oxid15 in #149
- Choose metric in HistoryViewer by @Oxid15 in #150
- View diffs in version logs by @Oxid15 in #151
- 0.10.0 by @Oxid15 in #152
- 0.10.0 - Update by @Oxid15 in #153
Full Changelog: v0.9.0...v0.10.0
v0.9.0 - Stability update
Version 0.9.0
New Cascade version is here - now it is an update that enhances stability and reliability of the package. New bugfixes, improved error messages, dataset versioning, more thorough testing.
Adds
- New way of dataset versioning. IMPORTANT! Be careful when updating from 0.8.0 because the way versioning works now changed and it will result in little bit inconsistent versions. #137 #138
skeleton
function to recursively extract the names of datasets from the pipeline #137- The notion of
SizedDataset
which should solve the inconsistencies with Dataset interface that has no__len__
method - Improved
ConstantBaseline
can now produce constants of any shape #135 - Metadata to
TimeSeriesDataset
#136
Refactors
- Copyright notices added where missing
- Better typing in all modules, parameters and methods
- Better code formatting
RangeSampler
now more protected against non-intended usagesplit
now more protected against missing parametersHistoryVIewer
's work with the linesMetricViewer.get_best_by
method is now informs that it does not work with non-sortable metricsBasicTrainer.train
protected against missingtest_ds
wheneval_strategy
providedYAMLHandler
now raises an error if reading an empty file
Fixes
- The problem in
UnderSampler
with repeating elements - The problem in
WeighedSampler
with string labels
0.8.0
Version 0.8.0
New Cascade version at last!
0.8.0 brings some useful changes and improvements
Adds
- Adds the ability to index lines in ModelRepo by integer indices #104
- Adds the ability to load best model by certain metric using MetricViewer #104
- The ability to customize saving and evaluation strategies in Trainer #107
- Check of hash in SkModel SkModel hash check #111
- Way to get dataset from Pickler #113
- ModelRepo automatically assigns name to a new line #114
- VersionAssigner - the way to automatically version datasets #112
- Composer #121
- WeighedSampler #125
- DataleakValidator #126
Refactors
- More stable start from checkpoint in Trainer #107
- Improves repo performance - now it loads much faster #109
- Work with meta lists #116
- Under- and OverSampler
Fixes
- MetaViewer behaviors #116
Deletes
- MetaHandlers read and write will be deprecated #109
- Removes the ability to use repo.lines as a public field making it protected #104
- Drops support of old meta files with no type keyword #110
- Drop SkModel name argument warning #110
- Drop SkClassifier #110
Docs
- New Trainer example #116
0.7.3 - Patch and new docs
0.7.2 - Patch and docs update
What's Changed
Full Changelog: v0.7.1...v0.7.2
0.7.1 - Patch
0.7.0
Version 0.7.0
Adds:
- The ability to save models without binaries in only_meta mode #83
- The ability to show only N last lines or models in HistoryViewer #85
- RangeSampler and split function #86
- The ability to F5 update in MetricViewer and HistoryViewers serve() method #87
- RandomSampler can now oversample #92
- Trainer and BasicTrainer classes #95
- The ability to pass a list of callables to Validator #99
Refactors:
- Protected fields #84
- Make more informative error messages in MetaHandlers #90
- Make MetaViewer lazy loaded, which should speed up large repos #90
- ModelRepos now are more reliable in case of missing or broken meta files #90
- Optimize ModelLine
- Refine error messages #91
- Make logging in repo more clear and reliable #91
- HistoryViewer can handle missing meta #93
- MetricViewer can handle missing meta #100
Fixes:
- Meta file's format can be chosen from ModelLine and ModelRepo #90
- Makes the work with ModelRepoConcatenator possible #94
- Enables the kwargs in serve method #94
Deletes:
Docs:
0.6.2 - Patch release
0.6.1 - Patch
0.6.0
What's Changed
- Dates are now dates and not just strings by @Oxid15 in #62
- Update meta from file by @Oxid15 in #63
- Use fixtures by @Oxid15 in #65
- Update tests by @Oxid15 in #64
- Drop skclassifier by @Oxid15 in #67
- Add TorchModel by @Oxid15 in #68
- Update docs by @Oxid15 in #69
- Model update by @Oxid15 in #70
- Abstract from json by @Oxid15 in #66
- Serve args by @Oxid15 in #71
- Debug num in mv by @Oxid15 in #73
- Concatenator's meta is now dict by @Oxid15 in #74
- TorchModel now has graph description in meta by @Oxid15 in #75
- Repo summing by @Oxid15 in #76
- Metric viewer update by @Oxid15 in #77
- Utils update by @Oxid15 in #78
- Docs update by @Oxid15 in #79
- 0.6.0 by @Oxid15 in #72
- 0.6.0 by @Oxid15 in #80
Full Changelog: v0.5.2...v0.6.0