huspacy-v0.9.0

Changed

• Added support for new models (hu_core_news_md-v3.5.2, hu_core_news_lg-v3.5.2, hu_core_news_trf_xl-v3.5.2, hu_core_news_trf_xl-v3.5.2)
• Updated documentation with benepar usage and the noun chunking

huspacy-v0.8.1

Fixed

• Replace bogus transformer model versions with fixed ones (hu_core_news_trf_xl-v3.5.1, hu_core_news_trf_xl-v3.5.1)

huspacy-v0.8.0

Fixed

• Applied an edit-tree lemmatizer fix, based on explosion/spaCy#12017

New

• Added support for new models (hu_core_news_md-v3.5.1, hu_core_news_lg-v3.5.1, hu_core_news_trf_xl-v3.5.0, hu_core_news_trf_xl-v3.5.1)

huspacy-v0.7.0

New

• Added support for new models (hu_core_news_md-v3.5.0, hu_core_news_lg-v3.5.0, hu_core_news_trf_xl-v3.4.0)
• Updated documentation

huspacy-v0.6.0

New

• Added a lookup component for sentiment lexicons
• Added integration for novakat's onpp NER model (nerpp)
• Added support for new models (hu_core_news_trf-v3.4.0, hu_core_news_md-v3.4.2, hu_core_news_lg-v3.4.4)

huspacy-v0.5.1

Bump version: 0.5.0 → 0.5.1

hu_core_ud_lg-0.3.1

Hungarian multi-task CNN trained on Universal Dependencies data. Assigns context-specific token vectors, Brown cluster IDs, word probabilities, POS tags, dependency parse, named entity tags and lemmata.

Feature	Description
Name	hu_core_ud_lg
Version	0.3.1
spaCy	>=2.2.1
Model size	1360 MB
Pipeline	tokenizer, sentencizer, tagger, parser, lemmatizer, ner
Vectors	1140008 unique vectors (300 dimensions)
Sources	Universal Dependencies, Szeged Corpus, Web Corpus, Wikipedia, Hunnerwiki, Szeged NER corpora
License	CC BY-NC-SA 4.0

Pipeline details

	Vectors	Tokenizer	Sentencizer	Tagger	Parser	Lemmatizer	NER
Model	Word2Vec CBOW dim=300 minfreq=10	Rule-based implemented in SpaCy	Rule-based	Multi-task CNN	Multi-task CNN	Lemmy (CST-like)	CNN
Training data	Wikipedia dump (2017-04-21)) and the Hungarian Webcorpus	-	-	CONLL'17 training data	CONLL'17 training data	UD converted Szeged Korpusz	Hunnerwiki, Szeged NER Business & Criminal
Test data	Hungarian analogical questions	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	Szeged NER Business & Criminal
Accuracy	ACC 20.95	F1 99.89	F1 96.97	ACC 94.81	UAS 76.18 LAS 66.58	ACC 95.51	F1 93.95

hu_core_ud_lg-0.3.0

Hungarian multi-task CNN trained on Universal Dependencies data. Assigns context-specific token vectors, Brown cluster IDs, word probabilities, POS tags, dependency parse, named entity tags and lemmata.

Feature	Description
Name	hu_core_ud_lg
Version	0.3.0
spaCy	>=2.1.8
Model size	1360 MB
Pipeline	tokenizer, sentencizer, tagger, parser, lemmatizer, ner
Vectors	1140008 unique vectors (300 dimensions)
Sources	Universal Dependencies, Szeged Corpus, Web Corpus, Wikipedia, Hunnerwiki, Szeged NER corpora
License	CC BY-NC-SA 4.0

Pipeline details

	Vectors	Tokenizer	Sentencizer	Tagger	Parser	Lemmatizer	NER
Model	Word2Vec CBOW dim=300 minfreq=10	Rule-based implemented in SpaCy	Rule-based	Multi-task CNN	Multi-task CNN	Lemmy (CST-like)	CNN
Training data	Wikipedia dump (2017-04-21)) and the Hungarian Webcorpus	-	-	CONLL'17 training data	CONLL'17 training data	UD converted Szeged Korpusz	Hunnerwiki, Szeged NER Business & Criminal
Test data	Hungarian analogical questions	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	Szeged NER Business & Criminal
Accuracy	ACC 20.95	F1 99.89	F1 96.97	ACC 94.91	UAS 75.73 LAS 66.16	ACC 95.49	F1 93.95

hu_core_ud_lg-0.2.0

Hungarian multi-task CNN trained on Universal Dependencies data. Assigns context-specific token vectors, Brown cluster IDs, word probabilities, POS tags, dependency parse and lemmata.

Feature	Description
Name	hu_core_ud_lg
Version	0.2.0
spaCy	>=2.1.0
Model size	1360 MB
Pipeline	tokenizer, sentencizer, tagger, parser, lemmatizer
Vectors	1140008 unique vectors (300 dimensions)
Sources	Universal Dependencies, Szeged Corpus, Web Corpus, Wikipedia
License	CC BY-NC-SA 4.0

Pipeline details

	Vectors	Tokenizer	Sentencizer	Tagger	Parser	Lemmatizer
Model	Word2Vec CBOW dim=300 minfreq=10	Rule-based implemented in SpaCy	Rule-based	Multi-task CNN	multi-task CNN	Lemmy (CST-like)
Training data	Wikipedia dump (2017-04-21)) and the Hungarian Webcorpus	-	-	CONLL'17 training data	CONLL'17 training data	UD converted Szeged Korpusz
Test data	Hungarian analogical questions	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data
Accuracy	ACC 20.95	F1 99.89	F1 96.97	ACC 94.82	UAS 78.02 LAS 67.92	ACC 95.60

hu_core_ud_lg-0.1.0

Hungarian multi-task CNN trained on Universal Dependencies data. Assigns context-specific token vectors, Brown cluster IDs, word probabilities, POS tags, dependency parse and lemmata.

Feature	Description
Name	hu_core_ud_lg
Version	0.1.0
spaCy	>=2.0.0
Model size	1350 MB
Pipeline	tokenizer, sentencizer, tagger, parser, lemmatizer
Vectors	1140008 unique vectors (300 dimensions)
Sources	Universal Dependencies, Szeged Corpus, Web Corpus, Wikipedia
License	CC BY-NC-SA 4.0

Pipeline details

	Vectors	Tokenizer	Sentencizer	Tagger	Parser	Lemmatizer
Model	Word2Vec CBOW dim=300 minfreq=10	Rule-based implemented in SpaCy	Rule-based	Multi-task CNN	multi-task CNN	Lemmy (CST-like)
Training data	Wikipedia dump (2017-04-21)) and the Hungarian Webcorpus	-	-	CONLL'17 training data	CONLL'17 training data	UD converted Szeged Korpusz
Test data	Hungarian analogical questions	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data	CONLL'17 test data
Accuracy	ACC 20.95	F1 99.88	F1 96.64	ACC 95.11	UAS 77.52 LAS 68.45	ACC 95.60