bot.zet PoS tagger

Bi-directional LSTM tagger with subword embeddings for EmpiriST/PostWITA datasets. The project now targets Python 3.12+, modern TensorFlow/Keras, and uses uv for dependency management.

Quickstart

1. Install dependencies:

   uv sync

2. Configure embedding paths in tagger.ini or via env vars:
   • TAGGER_W2V_SMALL=/path/to/small.vec
   • TAGGER_W2V_BIG=/path/to/big.vec
3. Train:

   uv run tagger train --task postwita --config tagger.ini --epochs 5

4. Predict:

   uv run tagger predict --model-path artifacts/tagger.keras --output-ext .pred

Project layout

• src/tagger/ package with config handling, data utilities, and the BiLSTM model.
• tagger.ini optional config; env vars prefixed with TAGGER_ override values.
• pyproject.toml defines runtime/dev dependencies; uv.lock produced by uv sync.

Notes

• Default data root is <repo>/data; override with --data-root or TAGGER_DATA_ROOT.
• Saved models use Keras’ .keras format containing architecture + weights.