Text-to-text Generation for Issue Report Classification

Submission for the NLBSE Issue Report Tool Competition 

This package accompanies the submission titled "Text-to-text Generation for Issue Report Classification" to the NLBSE Issue Report Tool Competition. The package provides resources for replicating the experiments and results presented.

Description of ZIP Files:

1. issue_classification_t5: This archive contains the code for replicating the study, including the retrieval of the pre-trained model, fine-tuning procedures, and inference execution.
   1. code: Contains all the code files.
      1. finetuning.py: The contents of this file comprise the code for fine-tuning the VMware/flan-t5-large-alpaca model on the issue report classification task. Additionally, embedded comments provide guidance on executing the fine-tuning process. Be sure to read the embedded comments.
      2. inference.py: This file contains the codebase for conducting inference using the fine-tuned model. Similar to the fine-tuning script, instructions for running the inference process are embedded as comments within the file.
      3. download_plm.py: This file contains the code for downloading VMware/flan-t5-large-alpaca from https://huggingface.co/VMware/flan-t5-large-alpaca . 
      4. requirements.txt: This file enumerates the required Python modules and their respective versions necessary for the successful execution of the provided code.
   2. data: Folder contains the NLBSE issue report classification data and model output after running inference using inference.py on issue-report-test.csv
      1. checkpoint-3000-output.csv: The contents of this CSV file present the output obtained after fine-tuning the VMware/flan-t5-large-alpaca model for 2 epochs (F1-score of 0.8297) on issue-report-train.csv and running the inference on issue-report-test.csv. Column 'label' contains the ground truth labels. Column 'Model generated output' contains the predicted label by the model.
      2. issue-report-train.csv: NLBSE24 isssue report tool competition train dataset. (Source: https://github.com/nlbse2024/issue-report-classification)
      3. issue-report-test.csv: NLBSE24 isssue report tool competition test dataset. (Source: https://github.com/nlbse2024/issue-report-classification)
2. finetuned_model_checkpoint-3000: This zip file contains the fine-tuned model (VMware/flan-t5-large-alpaca) to 2 epochs.