README.md

Inference for RaLEs Tasks

This directory provides scripts and utilities to perform inference (predictions) on the RaLEs tasks using trained models.

Key Files & Descriptions:

1. inference.py: used for Stanza NER dataset.

   • Usage: python inference.py --model_path [path_to_trained_model] --dataset_name [name_of_dataset] --data_split [train_val_test] --output_type [score_or_logits] --output_dir [path_to_inference_results_output_file]

2. [inference_procedure_2.py]: used for MIMIC III protocoling dataset.

   • Usage: python inference.py --model_path [path_to_trained_model] --dataset_name [name_of_dataset] --data_split [train_val_test] --output_type [score_or_logits] --output_dir [path_to_inference_results_output_file]

3. scripts directory containing scripts used to evaluate best trained models for the original RaLEs benchmark.

4. radiologygpt Script for evaluating RadiologyGPT, see more details in here

Step-by-step Guide:

1. Model Preparation: Ensure you have a trained model ready for inference. This could be a model you trained using the fine_tuning directory or a pre-trained model compatible with the RaLEs tasks.

2. Data Preparation: Make sure your input data for inference is in the correct format as expected by the inference scripts. Remember you can find the appropriate data preparation instructions here.

3. Perform Inference: Use the provided scripts as described above to make predictions on your datasets of interest. The scripts will generate outputs (predictions) which can be further processed or analyzed as needed.