songinpyo / used-car-price-regression-dacon

Predict used car price with car information. This is competition held in DACON

Used-Car-Price-Regression-DACON

Predict used car price with car information. This is competition held in DACON

Competition link : You can get Dataset from this link

https://dacon.io/competitions/official/235901/overview/description

Key Idea

Preprocessing key idea was transforming 'object' data to 'int' data.

Predict 'target' data with 'Pytorch DNN' and 'Pycaret Auto ML'

Tuning hyper parameter with 'Optuna'

Data set

Train data has 'title', 'odometer', 'location', 'isimported', 'engine', 'paint', 'year', 'target' columns

Test data has 'title', 'odometer', 'location', 'isimported', 'engine', 'paint', 'year' columns.

Code example

models = [
    ('LGBMRegressor',LGBMRegressor(n_estimators=800,max_depth=13, learning_rate=0.01,subsample=0.7,random_state=0)),
    ('XGBRegressor',XGBRegressor(n_estimators=1600,max_depth=13,learning_rate=0.02,random_state=0,subsample=0.7,verbosity=0)),
    ('CatBoostRegressor',CatBoostRegressor(n_estimators=800,max_depth=10,learning_rate=0.02,subsample=0.8,random_state=0,verbose=0)),
    ('GradientBoostingRegressor',GradientBoostingRegressor(n_estimators=1200,max_depth=9,learning_rate=0.01,subsample=0.8,random_state=0)),
    ('ExtraTreesRegressor',ExtraTreesRegressor(n_estimators=500,max_depth=9,random_state=0,criterion='mse')),
    ('RandomForestRegressor',RandomForestRegressor(n_estimators=4000,max_depth=9,random_state=0,criterion='mse'))]
    
for name, model in models:
    model.fit(x_train, y_train)
    print(f'{name}: ', NMAE(y_val,model.predict(x_val)))

Future work

It could be better to change 'odometer' int data into categorical data.

Example) from 0 to 10000 = 0, from 10000 to 20000 = 1 ...