| | --- |
| | license: mit |
| | tags: |
| | - TensorFlow |
| | - BERT |
| | - Transformer |
| | - Classification |
| | - Regression |
| | --- |
| | # Code Qualiy Evaluation Dataset |
| | Welcome to the repository for our research paper: T. Wang and Z. Chen, "Analyzing Code Text Strings for Code Evaluation," 2023 IEEE International Conference on Big Data (BigData), Sorrento, Italy, 2023, pp. 5619-5628, doi: 10.1109/BigData59044.2023.10386406. |
| |
|
| | ## Contents |
| | This repository contains the following: |
| | - Fine-tuned Model |
| | - Dataset (https://github.com/tisage/codeQuality) |
| | - License |
| |
|
| | ## Model Info |
| | There are three BERT models, each fine-tuned on a dataset of 70K Python 3 solutions submitted by users for problems #1 through #100 on LeetCode: |
| | - `bert_lc100_hp25`: This model classifies code based on the 25th percentile as its threshold. It is designed for identifying lower quartile code solutions in terms of quality or performance. |
| | - `bert_lc100_hp50`: Operating with a median-based approach, this model uses the 50th percentile as its classification threshold. It is suitable for general assessments, providing a balanced view of code quality. |
| | - `bert_lc100_regression`: Unlike the others, this is a regression model. It provides a nuanced prediction of the overall code quality score, offering a more detailed evaluation compared to the binary classification approach. |
| | - `bert_lc100_regression_v2`: similar to `bert_lc100_regression` model, the correctness score is calculated using more restricted rule `==` instead of similarity. |
| |
|
| | ## Model Usage |
| | **Installation** |
| | First, ensure you have the latest version of the tf-models-official package. You can install it using the following command: |
| | ``` |
| | pip install -q tf-models-official |
| | ``` |
| |
|
| | **Loading the Model** |
| | To utilize the bert_lc100_regression model within TensorFlow, follow these steps: |
| | ``` |
| | import tensorflow as tf |
| | import tensorflow_text as text |
| | model = tf.keras.models.load_model('saved_model/bert_lc100_regression/', compile=False) |
| | ``` |
| |
|
| | **Making Predictions** |
| | To assess the quality of code, given that `X_test` contains a list of code strings, use the model to predict as follows: |
| | ``` |
| | y_pred = model.predict(X_test) |
| | ``` |
| |
|
| | ## Reference |
| | If you found the dataset useful in your research or applications, please cite using the following BibTeX: |
| | ``` |
| | @INPROCEEDINGS{10386406, |
| | author={Wang, Tianyu and Chen, Zhixiong}, |
| | booktitle={2023 IEEE International Conference on Big Data (BigData)}, |
| | title={Analyzing Code Text Strings for Code Evaluation}, |
| | year={2023}, |
| | volume={}, |
| | number={}, |
| | pages={5619-5628}, |
| | keywords={Measurement;Deep learning;Codes;Bidirectional control;Organizations;Transformers;Software;code assessment;code annotation;deep learning;nature language processing;software assurance;code security}, |
| | doi={10.1109/BigData59044.2023.10386406} |
| | } |
| | ``` |
