Blood-Spectroscopy-Classification-using-Deep-Learning

A starter solution code(including the best final solution iteration) to a machine learning problem from Zindi.Africa.

Find link to final notebook here

Starter iteration utilized FFT(fast fourier transform) features to train a neural network architecture in a single/multi-input fashion.

The ultimate goal of this project was to accurately identify the levels of different blood-based substrates from Near-Infrared Reflectance(NIR) spectral data.

Full description of challenge and datasets acquisition can be accessed here

How to ran starter scripts in single/multi-input modes

single input:

python train.py --train_csv Train.csv --test_csv Test.csv --use_threshold --use_smoothing --BATCH_SIZE 32
--EARLY_STOP --EPOCHS 100 --WEIGHT_DECAY 1e-6 --model_type single --model_name NN

multiple inputs:

python train.py --train_csv Train.csv --test_csv Test.csv --use_threshold --use_smoothing --BATCH_SIZE 32
--EARLY_STOP --EPOCHS 100 --WEIGHT_DECAY 1e-6 --model_type double --model_name NN

run inference

python make_submission.py --model_name NN --submission_dir path/to/submissions

Leaderboard

Rank : 14/265

Authors

Name	Zindi ID	Github ID
Ronny Polle	@100i	@DrCod