Directory contains additional data bundled with the Nerf tool.

Schemes directory
=================

In the schemes directory you will find examplary schema files.
Schema final.cfg has been used to train the final CRF model
(the model can be found in the models directory).

Models directory
================

Here you can find the final CRF model trained on the 1.0 version
of 1-milion, manually annotated NKJP subcorpus. The following
command has been used to produce the final model:

  python scripts/train.py --schema=../data/schemes/final.cfg --tei-corpus=PATH/TO/NKJP-PodkorpusMilionowy-1.0 --model-out=../data/models/final.tgz --pre-train --iter-num=100 --threads=10 --scale0=0.05 --relative-eval-size=0.0

In the models directory there is also an atlas.tgz model, which
has been trained on the same corpus but with simpler schema and
with changed NEs hierarchy (no nested or discontinuous entities,
no subtypes, relational NEs removed, additional MONEY and
PERCENTAGE types).
