Evaluation: Implement evaluation

[ybracke]

• See Bawden et al. (2022) (Zotero, Github) for an interesting evaluation metric
• See Robertson & Goldwater (2018) (Zotero)
• See normeval repo
• Integrate evaluation in dvc exp pipeline (see #71)

[ybracke]

Sketch for evaluation process

1. Generating predictions

• [x] Add generate.py to generate normalizations. This script takes a config file specifying the model, the test parameters and the test data.

2. Compute evaluation metrics

• [x] The to-be-named evaluation module is installed via pip and offers an entry-point to call the evaluation on a given JSONL file is copied to the evaluate directory and can be applied to the JSONL file created by generate.py

Evaluation returns a JSONL file with metrics. Example output:

{"n": 100,  "acc_harmonized": 0.95,  "dist_normalized": 0.1} 

:white_check_mark: Done