Gradient Boosting Decision Tree Training using PSO

Packages Required

In order to run the main program, you need to install the following packages.

These packages are:

• Numpy
• Pandas
• XGBoost
• Sklearn
• CatBoost
• pickle
• matplotlib

Experiment

Use the following command to run experiment for both SGBDT and PSGBDT.

Run SGBDT by python run.py

You can specify the dataset path by --dataset_path $dataset name , currelty support

• BankNote.csv (classification)
• wine.csv (classification)
• winequality-red.csv (regression)
• higgs_0.005.csv (classification)
• covat_0.3.csv (classification)
• insurance.csv (regression)
• kc_house_data.csv (regression)

and --model_type need to be specified by

• binary_cf
• regressiom

with respect to the type of dataset

Run PSGBDT by specifying the --pretrain_file and --pretrain_type, pretrain files are under the folder pretrain_models. And --pretrain_type needs to set as either xgb or skr

Example

Run SGBDT:

python run.py --dataset_path kc_house_data.csv --model_type regression

Run PSGBDT:

python run.py --dataset_path kc_house_data.csv --model_type regression --pretrain_file kc_house.pkl --pretrain_type xgb