Logo Lanfrica

truong2710-cyber/AI7101-Group-Assignment

Domain:

environment and energy
Creator:
tru
Host:
African Air Quality Prediction Project for AI7101 @ MBZUAI # African Air Quality Prediction Final project - AI7101: Machine Learning with Python @ MBZUAI, Fall 2025 [Slides | Report] Competition on Zindi ## Installing dependencies `pip install -r requirements.txt` ## File structure `dataset`: contains the train and test sets in 2 .csv files. `eda`: contains a notebook for EDA, running will create a new folder named `imgs` to store the result figures. `src`: main structure of the project. `output`: stores .csv files per run for submission. `solution.py`: main file for execution of the whole ML pipeline, including preparing a file for submission. ## Script for training, validation, and producing test file ```bash python solution.py \ --exp_name \ --missing_threshold \ --top_features \ --corr_threshold \ --clip_threshold \ --models \ --feature_selection_method \ --drop_location \ --augment_date \ --use_unify \ --use_cloud_diff \ --scale_target \ --clip_target ``` Example ```bash python solution.py \ --exp_name baseline_pm25 \ --missing_threshold 1.0 \ --top_features 40 \ --corr_threshold 0.9 \ --clip_threshold 0.97 \ --models cat,lgb,xgb,lasso,svr \ --feature_selection_method catboost \ --drop_location False \ --augment_date True \ --use_unify False \ --use_cloud_diff False \ --scale_target False \ --clip_target True ``` Explanation of variables: `--exp_name`: name of your experiment to be saved `--missing_threshold`: threshold to eliminate null columns `--top_features`: number of features kept after feature selection `--corr_threshold`: correlation threshold to eliminate features `--clip_threshold`: target variable quantile to be clipped, used with `--clip_target` `--models`: list of models to be ensembled, a comma seperated list such as 'cat,lgb,xgb,lasso,svr' `--feature_selection_method`: one of 'catboost','anova','lasso','permutation','all' `--drop_location`, `--augment_date`, `--use_unify`, `--use_cloud_diff`: feature engineering methods `--scale_target`: whether to log-scale the …