Below you can find a outline of how to reproduce my solution for the azubi africa competition. If you run into any trouble with the setup/code or have any questions please contact me us at [email protected]

#ARCHIVE CONTENTS model.rar : original model upload - contains original code, additional training examples, corrected labels, etc comp_etc : contains ancillary information for prediction - clustering of training/test examples comp_mdl : model binaries used in generating solution comp_preds : model predictions train_code : code to rebuild models from scratch predict_code : code to generate predictions from model binaries

#HARDWARE: (The following specs were used to create the original solution) Edition Windows 10 Pro - Version 2004 - OS build 19041.450 (256 GB boot disk) 8 core CPU, 16 GB memory 1 x NVIDIA 940mx

#SOFTWARE (python packages are detailed separately in requirements.txt): Python 3.7.8 CUDA 11.0.228 nvidia drivers v.3.20.4.14

#DATA SETUP ./Azubi Africa/data/raw/stage1/label_data.csv

#DATA PROCESSING #The train/predict code will also call this script if it has not already been run on the relevant data. python ./train_code/prepare_data.py --data_dir=data/stage1/ --output_dir=data/stage1_cleaned

#MODEL BUILD This model is build with the MaxAbsScaler and AdaBoost Classifier The prediction runs in a few minutes and uses precomputed model predictions. Saved model binaries stored in Azubi Afirca/submissions/comp_mdl