steinmann / human-classifier Goto Github PK

• a simple command line interfacte to feed data, return model accuracy and save the model
• optional tuning mode performing randomized search
• uses tuned xgboost binary classifier
• excludes heroes that lack info on race and superpowers
• uses all properties except name as features
• categorical attributes are one hot encoded
• uses five fold cross validation to determine the model accuracy

usage: train.py [-h] --info [INFO] --power [POWER] [--model [MODEL]]

-h, --help show this help message and exit

--info [INFO] csv file containing general superhero info

--power [POWER] csv file containing info on superhero powers

--model [MODEL] name of the model (optional)

--tune freshly tune parameters using random search