Improve funding mechanism using XGBoost and BART LLM

Deepfunding competition

for full description about the competition please refer to cryptopond platform https://cryptopond.xyz/modelfactory/detail/306250?tab=0

Huggingface : https://huggingface.co/spaces/DeepFunding/PredictiveFundingChallengeforOpenSourceDependencies

Workflow

(in this repo you just need to run main_huggingface_XGB.ipynb, i stored all csv helper.)

1. Data Collection

• Download train and test data from both platform
• Create a list contains all github repo urls (all_github_urls.csv)
• For each url in all_github_urls, use get_metrics.ipynb to get repository metrics.
• Scrape frontend data on each repo link to get closed/open issues, open/closed pull-request, and used by number by running scrape_frontend_github_repo.ipynb.
• Using summarize_readme.ipynb Get readme file on each repo.

2. Data preprocessing

• extract repository name
• apply log transformations and ratio for numerical features like stars, watchers, etc.
• convert boolean values to integer
• encode categorical variables
• using BART to create summarization of repo's readme file
• TF-IDF Vectorization
• feature selection

3. Model Training

• train XGBoost model using preprocessed data
• evaluate the model (MSE and R^2)

4. Making predictions

• Make final predictions on the test dataset
• create csv submission file
• transitivity check

honorable mention :

@davidgasques (metrics & ratio-ing) https://github.com/davidgasquez/predictive-funding-challenge

@Faezehshakouri https://github.com/FaezehShakouri/deepfunding