Evaluation: Code for analysis

[ybracke]

• Generations (or "predictions") were previously created with this Jupyter notebook: notebooks/exploratory/inspect_predictions.ipynb (now only locally on gpu server: hidden/trash/inspect_predictions.ipynb
• The notebook contains some analysis supporting functions. They could be reused.
• Perhaps we want a notebook or something that reads in JSONL files with fields like "orig", "gold" and "pred" and applies the analysis functions.
• Could Meld be helpful for a visual comparison of orig, gold and pred?