Stylometry for Real-World Expert Coders: a Zero-shot Approach

This release contains the replication artefacts regarding the experiments present in the work: "Stylometry for Real-World Expert Coders: a Zero-shot Approach". It encompasses the replication package containing the code used during the experiments, both in-distribution and out-distribution datasets associated with the research paper and the three checkpoints related to the trained models. The 'replication_package' zip file contains a README file with the instructions to replicate the experiments, the requirements file to reproduce the coding environment used during the experiments and two subfolders, the 'CCS' with the snippets related to the stylometry model experiments and the 'dataMining' with the snippets related to the data mining phase. The file README.md is a copy of the instructions contained in the replication package zip and it includes the instructions to replicate the experiments. The 'inDistributionSet' zip contains the dataset mined and split used for the training and in-distribution testing procedure, it follows the same experimental setup as the train, validation, and test sets, whereas the 'outDistributionSet' zip, exclusively comprises snippets used for testing purposes. The 'models checkpoints' zip file contains:- MLAllVocaBSoftAtt referring to the ML model model checkpoint trained with infoNCE loss without B.P.E, with bounding.- SoftAtt64k104AuthClass referring to the softAttn-classifier model checkpoint trained in a classification setup with 64k B.P.E. token, with bounding.- SelfAtt64k104authClass referring to the selfAttn-classifier model checkpoint with 64k B.P.E. token, without bounding.