AlukweClaid/AlukweClaid-week-7-zindi-challenge-swahili-news

# Swahili News Classification Challenge This repository contains data and modeling assets for classifying Swahili news articles into five categories: `kitaifa`, `kimataifa`, `biashara`, `michezo`, and `burudani`. ## Dataset - Training data: 5,151 labeled news articles in `Train.csv` - Test data: 1,030 unlabeled news articles in `Test.csv` - Data dictionary: `VariableDefinitions.csv` ## Project Structure ```text |- group_14_EDA_week_7.ipynb |- requirements.txt |- Train.csv |- Test.csv |- SampleSubmission.csv `- results/ ``` ## Recommended Run Path The notebook is now the main reproducible workflow. It: - validates the input files - normalizes label casing to lowercase - trains a TF-IDF + balanced Logistic Regression baseline - evaluates on a stratified validation split - writes output files to `results/` - includes improved prompt-engineering utilities for optional LLM-assisted review Generated files: - `results/baseline_metrics_from_notebook.json` - `results/baseline_submission_from_notebook.csv` ## Installation ```bash pip install -r requirements.txt ``` This installs the notebook dependencies for: - baseline modeling - visualizations - optional OpenAI API usage for prompt-engineering experiments - XGBoost and multilingual BERT sections If you only want the lightest baseline environment, the minimum practical packages are `pandas`, `numpy`, and `scikit-learn`. ## Notes - The raw labels in `Train.csv` are inconsistently cased, so normalization is required for stable training and submission output. - `results/checkpoint-*` directories come from earlier notebook training runs and are not required for the notebook baseline. ## License This repository is for a challenge project and does not include a formal license file.