=================================== TRAINING ===================================

training.py

-> call with -h or --help for help
-> more options available by modifying settings.py

+------------------------------------------------------------------------------+
| The Keras library used by the script can be greedy on CPU cores.             |
| Wrap the call e.g. in taskset to restrict the number of cores used.          |
+------------------------------------------------------------------------------+

TRAINING DATA FORMAT:
* Tab-separated file with following columns:
    - wordform
    - possible morphological analyses, pipe-separated (e.g. subst:sg:inst:n|ger:sg:inst:n:perf:aff)
    - correct morphological analysis
    - token label (may be left empty except for IGN for unknown words)
    - (unused)
    - (unused)

Use PolEval/prepare_data.py to convert XCES to this format.
Use dag2training.py to convert Morfeusz/Concraft DAG to this format.

Define your tagset by modifying tagset.py.

REQUIRED RESOURCES:
* If Concraft2 input is to be included in input features:
    - Concraft2 binary (set CONCRAFT_BIN in concraft_utils.py accordingly)
    - tagset config file
    - a directory where created Concraft data will be placed (DATA_DIR in concraft_utils.py)
* If Word2Vec embeddings are to be included in input features:
    - pretained models created by Aleksander Wawer, available from http://mozart.ipipan.waw.pl/~axw/models/orth/ (set WORD2VEC_MODELS_DIR is data.py to the directory where the models reside, do not change the model files’ names)

=================================== TAGGING ====================================

tagging.py
-> call with -h or --help for help

CURRENTLY SUPPORTED INPUT FORMATS:
* XCES