/SE3M

This repository contains the source codes and data set used in the experiments at the article entitled "SE3M: A model for estimating software effort using pre-trained embeddings models".

Primary LanguageJupyter Notebook

SE3M

This repository contains the source codes and data set used in the experiments at the article entitled "SE3M: A model for estimating software effort using pre-trained embeddings models" (FÁVERO et al., 2020).

Fávero, E. M., Casanova, D., & Pimentel, A. R. (2020). SE3M: A Model for Software Effort Estimation Using Pre-trained Embedding Models. arXiv preprint arXiv:2006.16831.

Other related works:

  • Fávero, E. M. D. B., Pereira, R., Pimentel, A. R., & Casanova, D. (2018). Analogy-based Effort Estimation: A Systematic Mapping of Literature. INFOCOMP Journal of Computer Science, 17(2), 07-22.

  • Fávero, E. M. D. B., Casanova, D., & Pimentel, A. R. (2019, September). EmbSE: A Word Embeddings Model Oriented Towards Software Engineering Domain. In Proceedings of the XXXIII Brazilian Symposium on Software Engineering (pp. 172-180).

Resources available:

  1. Data set (user story) labeled [1], used for training and test ing the inference model. https://github.com/morakotch/datasets/tree/master/storypoint/IEEE%20TSE2018/dataset

    • Correspond to a set of .CSV files for each of the projects used.

  2. Pre-trained embeddings (generic).

    • Available in the folder "pretrain_model"
      • word2vec_base
      • BERT_base

  3. Unlabeled data set (user story) used in the fine-tuning process of pre-trained embeddings. https://github.com/morakotch/datasets/tree/master/storypoint/IEEE%20TSE2018/pretrain%20data

  4. The pre-processing of the data used to perform the fine-tuning process with BERT, as well as fine-tuning, used the methods provided by the BERT model in its official repository at https://github.com/google-research/bert

    • For data pre-processing: create_pretraining_data.py
      • Standard parameters were used, changing only the following:: -input_file= (inform file .txt containing all the textual requirements provided in item 2) -output_file=./filename.tfrecord -vocab_file= (inform file .txt corresponding to the vocabulary of the pre-trained model used, ex./uncased_L-12/vocab.txt)

  5. Pre-trained embeddings (fine-tuned) models for the specific domain of software engineering (SE):

    • word2vec_SE
    • BERT_SE

  6. The "SE3M_model.ipynb" file contains a deep learning of architecture used as an inference model for estimating software effort by analogy. Is a Google Colab notebook, simply replacing the paths of the files used.

References:

[1] M. Choetkiertikul, HK Dam, T. Tran, TTM Pham, A. Ghose e T. Menzies, "A deep learning model for estimating story points.", IEEE Trans. Softw. Eng. Vol. PP, não. 99, p. 1, 2018.