jeannefukumaru / nb_dep_ud_sm

Spacy model trained based on Norwegian corpus converted from OBT to Universal dep.

Experimental Norwegian (Bokmål) language model for Spacy

This model is based of the Norwegian Universal dependency dataset that can be found here:

https://github.com/UniversalDependencies/UD_Norwegian-Bokmaal

Command used to train the model:

batch_from=16 batch_to=64 python -m spacy train nb model_out no_bokmaal-ud-train.json no_bokmaal-ud-dev.json -n 30

There is probably much room for improvement on this model. However in regards to tagging the model seems to perform pretty well.

Iteration 7 seemed to be working best so this is the one packaged here.

To get the same results as show here please use the updated Norwegian language package for Spacy. It should now be a part of the master branch, but the Pull request can be found here: explosion/spaCy#1882

Installation

To install the package use this command:

pip install https://github.com/ohenrik/nb_dep_ud_sm/raw/master/nb_dep_ud_sm-0.0.1/dist/nb_dep_ud_sm-0.0.1.tar.gz

Usage

import spacy
nb = spacy.load("nb_dep_ud_sm")

doc = nb("Det er kaldt på vinteren i Norge.")

Training results:

Itn.	P.Loss	N.Loss	UAS	NER P.	NER R.	NER F.	Tag %	Token %	na	na
0	500.962	0.000	83,67	0.000	0.000	0.000	93,269	100.000	3542.9	0.0
1	86.554	0.000	86,38	0.000	0.000	0.000	94,396	100.000	3767.6	0.0
2	35.351	0.000	87,07	0.000	0.000	0.000	94,762	100.000	3611.1	0.0
3	21.769	0.000	87,99	0.000	0.000	0.000	94,839	100.000	3779.8	0.0
4	19.490	0.000	88,26	0.000	0.000	0.000	95,02	100.000	3565.9	0.0
5	17.730	0.000	88,48	0.000	0.000	0.000	95,084	100.000	3421.0	0.0
6	16.141	0.000	88,77	0.000	0.000	0.000	95,042	100.000	3533.3	0.0
7	14.906	0.000	88,72	0.000	0.000	0.000	95,139	100.000	3572.3	0.0
8	13.644	0.000	88,76	0.000	0.000	0.000	95,042	100.000	3585.8	0.0
9	12.909	0.000	88,72	0.000	0.000	0.000	95,125	100.000	3694.2	0.0
10	12.194	0.000	88,72	0.000	0.000	0.000	95,075	100.000	3618.3	0.0
11	11.435	0.000	88,65	0.000	0.000	0.000	95,042	100.000	3738.2	0.0
12	10.950	0.000	88,67	0.000	0.000	0.000	94,754	100.000	3909.9	0.0
13	10.325	0.000	88,85	0.000	0.000	0.000	47,879	100.000	3673.9	0.0
14	9.793	0.000	88,88	0.000	0.000	0.000	42,063	100.000	3758.4	0.0
15	9.456	0.000	88,77	0.000	0.000	0.000	43,68	100.000	3497.1	0.0
16	8.967	0.000	88,69	0.000	0.000	0.000	45,06	100.000	3514.9	0.0
17	8.493	0.000	88,88	0.000	0.000	0.000	46,537	100.000	3632.7	0.0
18	8.109	0.000	88,76	0.000	0.000	0.000	47,249	100.000	3837.6	0.0
19	7.795	0.000	88,73	0.000	0.000	0.000	47,485	100.000	3473.2	0.0
20	7.573	0.000	88,81	0.000	0.000	0.000	47,579	100.000	3482.8	0.0
21	7.131	0.000	88,82	0.000	0.000	0.000	47,282	100.000	3327.1	0.0
22	7.053	0.000	88,87	0.000	0.000	0.000	46,916	100.000	3576.0	0.0
23	6.736	0.000	88,61	0.000	0.000	0.000	46,394	100.000	3223.6	0.0
24	6.459	0.000	88,83	0.000	0.000	0.000	45,841	100.000	3523.7	0.0
25	6.364	0.000	88,67	0.000	0.000	0.000	45,423	100.000	3163.7	0.0
26	6.080	0.000	88,80	0.000	0.000	0.000	44,959	100.000	3497.2	0.0
27	5.984	0.000	88,77	0.000	0.000	0.000	44,56	100.000	3642.3	0.0
28	5.724	0.000	88,99	0.000	0.000	0.000	44,249	100.000	3467.4	0.0
29	5.620	0.000	88,97	0.000	0.000	0.000	43,895	100.000	3628.4	0.0

Not an official model

This is not yet an official spacy model