All notable changes to this project will be documented in this file.
- Add intents filter parameter in parsing CLI #858
- Add documentation about intents filters #858
- Update dependencies for better python3.8 support #867
0.20.1 - 2019-09-04
- Allow to bypass the model version check #830
- Persist
CustomEntityParserlicense when needed #832 - Document metrics CLI #839
- Allow to fit SnipsNLUEngine with a
Datasetobject #840
- Update
snips-nlu-parsersdependency upper bound to 0.5 #850
- Invalidate importlib caches after dynamically installing module #838
- Automatically generate documentation for supported languages and builtin entities #841
- Fix issue when cleaning up crfsuite files #843
- Fix filemode of persisted crfsuite files #844
0.20.0 - 2019-07-16
- Add new intent parser:
LookupIntentParser#759
- Replace
DeterministicIntentParserbyLookupIntentParserin default configs #829 - Bumped
snips-nlu-parsersto0.3.xintroducing new builtin entities:snips/timesnips/timePeriodsnips/datesnips/datePeriodsnips/citysnips/countrysnips/region
0.19.8 - 2019-07-10
- Add filter for entity match feature #814
- Add noise re-weight factor in
LogRegIntentClassifier#815 - Add warning logs and improve errors #821
- Add random seed parameter in training CLI #819
- Fix non-deterministic behavior #817
- Import modules lazily to speed up CLI startup time #819
- Removed dependency on
semantic_versionto accept"subpatches"number #825
0.19.7 - 2019-06-20
- Re-score ambiguous
DeterministicIntentParserresults based on slots #791 - Accept ambiguous results from
DeterministicIntentParserwhen confidence score is above 0.5 #797 - Avoid generating number variations when not needed #799
- Moved the NLU random state from the config to the shared resources #801
- Reduce custom entity parser footprint in training time #804
- Bumped
scikit-learnto>=0.21,<0.22forpython>=3.5and>=0.20<0.21forpython<3.5#801 - Update dependencies #811
- Fixed a couple of bugs in the data augmentation which were making the NLU training non-deterministic #801
- Remove deprecated code in dataset generation #803
- Fix possible override of entity values when generating variations #808
0.19.6 - 2019-04-26
- Raise an error when using unknown intents in intents filter #788
- Fix issue with stop words in
DeterministicIntentParser#789
0.19.5 - 2019-04-10
- Advanced inference logging in the
CRFSlotFiller#776 - Improved failed linking error message after download of resources #774
- Improve handling of ambiguous utterances in DeterministicIntentParser #773
- Remove normalization of confidence scores in intent classification #782
- Fixed a crash due to missing resources when refitting the
CRFSlotFiller#771 - Fixed issue with egg fragments in download cli #769
- Fixed an issue causing the
Noneintent to be ignored when using theparseAPI in conjunction withintentsandtop_n#781
0.19.4 - 2019-03-06
- Support for Portuguese: "pt_pt" and "pt_br"
- Enhancement: leverage entity scopes of each intent in deterministic intent parser
0.19.3 - 2019-03-05
- Issue with intent classification reducing classification accuracy
- Issue resulting in a mutation of the CRFSlotFillerConfig
- Wrong required resources of the
DeterministicIntentParser - Issue with non ASCII characters when using the parsing CLI with Python2
0.19.2 - 2019-02-11
- Fix an issue regarding the way builtin entities were handled by the
CRFSlotFiller
0.19.1 - 2019-02-04
- Bug causing an unnecessary reloading of shared resources
0.19.0 - 2019-02-04
- Support for Python3.7
get_intents(text)API inSnipsNLUEngineto get the probabilities of all the intentsget_slots(text, intent)API inSnipsNLUEngineto extract slots when the intent is known- The
DeterministicIntentParsercan now ignore stop words through the newignore_stop_wordsconfiguration parameter - Co-occurrence features can now be used in the
LogRegIntentClassifier
- Remove the deprecated text file format for intents and entities in favor YAML format
- The
Noneintent is now handled as a regular intent in the parsing output, which means that:
{
"input": "foo bar",
"intent": None,
"slots": None
}is replaced with:
{
"input": "foo bar",
"intent": {
"intentName": None,
"probability": 0.552122
},
"slots": []
}- Patterns of the
DeterministicIntentParserare now deduplicated across intents in order to reduce ambiguity - Improve the use of custom
ProcessingUnitthrough the use ofRegistrablepattern - Improve the use of default processing unit configurations
- Improve logging
- Replace
snips-nlu-ontologywithsnips-nlu-parsers - Drop support for Python3.4
- Issue when persisting resources
- Issue when resolving custom entities
- Issue with whitespaces when generating dataset from YAML and text files
- Issue with unicode when using the CLI (Python 2)
0.18.0 - 2018-11-26
- New YAML format to create dataset
- Verbose mode in CLI
- Bump
snips-nlu-ontologyto0.62.0to improve memory usage
0.17.4 - 2018-11-20
- Add a
--configargument in the metrics CLI
- Replace "parser_threshold" by "matching_strictness" in dataset format
- Optimize loading and inference runtime
- Disable stemming for intent classification in default configs
0.17.3 - 2018-10-18
- Crash with num2words and floats
0.17.2 - 2018-10-15
- Support for builtin music entities in english
0.17.1 - 2018-10-09
DeterministicIntentParsernow relies on the custom entity parser
- Bump
snips-nlu-ontologyto0.60
0.17.0 - 2018-10-05
- Support for 3 new builtin entities in French:
snips/musicAlbum,snips/musicArtistandsnips/musicTrack - Minimal support for Italian
- model version
0.16.0=>0.17.0
- Bug with entity feature name in intent classification
0.16.5 - 2018-09-06
- Segfault in CRFSuite when the
CRFSlotFilleris fitted only on empty utterances
0.16.4 - 2018-08-30
- Issue with the
CrfSlotFillerfile names in theProbabilisticIntentParserserialization
0.16.3 - 2018-08-22
- Issue with synonyms when multiple synonyms have the same normalization
0.16.2 - 2018-08-08
automatically_extensibleflag in dataset generation tool- System requirements
- Reference to chatito tool in documentation
- Bump
snips-nlu-ontologyto0.57.3 - versions of dependencies are now defined more loosely
- Issue with synonyms mapping
- Issue with
snips-nlu download-all-languagesCLI command
0.16.1 - 2018-07-23
- Every processing unit can be persisted into (and loaded from) a
bytearray
0.16.0 - 2018-07-17
- The
SnipsNLUEngineobject is now persisted to (and loaded from) a directory, instead of a single json file. - The language resources are now persisted along with the
SnipsNLUEngine, removing the need to download and load the resources when loading a trained engine. - The format of language resources has been optimized.
- Stemmed gazetteers, computed beforehand. It removes the need to stem gazetteers on the fly.
- API to persist (and load) a
SnipsNLUEngineobject as abytearray
- Issue in the
DeterministicIntentParserwhen the same slot name was used in multiple intents while referring to different entities
0.15.1 - 2018-07-09
- Bump
snips-nlu-ontologyto0.57.1
- Crash when parsing implicit years before 1970
0.15.0 - 2018-06-21
- Language resources are now packaged separately from the Snips NLU core
library, and can be fetched using
snips-nlu download <language>. - The CLI tool now consists in a single entry point,
snips-nlu, which exposes several commands.
- CLI command to parse a query
0.14.0 - 2018-06-08
- Issue due to caching of builtin entities at inference time
- Improve builtin entities handling during intent classification
- Improve builtin entities handling in
DeterministicIntentParser - Reduce size of regex patterns in trained model file
- Update model version to
0.15.0
0.13.5 - 2018-05-23
- Fixed synonyms matching by using the normalized version of the tagged values
- Fixed dataset augmentation by keeping stripped values of entities
- Fixed the string variations functions to prevent generating too many variations
0.13.4 - 2018-05-18
- Documentation for the
Noneintent
- Improve calibration of intent classification probabilities
- Update snips-nlu-ontology version to 0.55.0
- DeterministicIntentParser: Fix bug when deduplicating regexes
- DeterministicIntentParser: Fix issue with incorrect ranges when parsing sentences with both builtin and custom slots
- DeterministicIntentParser: Fix issue with builtin entities placeholders causing mismatches
- Fix issue with engine-inference CLI script not loading resources correctly
0.13.3 - 2018-04-24
- Add config parameter to augment data with builtin entity examples
- Bump snips-nlu-ontology to 0.54.3
- Use language specific configs by default
- Add right space to chunks in data augmentation
- Update JA config
- Fix inconsistency bug with shape ngram CRF feature
- Fix bug when initializing
CRFSlotFillerwith config dict - Fix bug with gazetteer in ngram feature
- Fix bug with length CRF feature
0.13.1 - 2018-04-10
- Bump ontology version from 0.54.1 to 0.54.2
- Fix CRF parsing of builtin entities by adding builtin entities examples of different length
- Fix CLI scripts importing metrics package which might not be installed
0.13.0 - 2018-04-06
- Add contributing guidelines, code of conduct, authors and contributors
- Add integration test
- Add CHANGELOG
- Bump model version from 0.13.0 to 0.14
- Improve intent classification by leveraging builtin entities
- Improve loading of language specific resources
- Improve support of japanese
- Remove
exhaustive_permutations_thresholdparameter in config
- Fix compiling issue with
bindgendependency when installing from source - Fix issue in
CRFSlotFillerwhen handling builtin entities