Tree Based Classification

A C++ implement of GBDT + Bagging Classification.

Data

Data for this program is available on Large-scale classification-SYSU-2017, where we can also find the description of this classification task.

Model Parameter

Modify model parameters in src/def.h

• BAGGING_SIZE: bagging size
• BOOSTING_SIZE: boosting size
• MAX_LEAVES: max leaves of one tree
• MAX_DEPTH: max depth of one tree
• DECAYING_RATE: decaying rate along the out put of boosting trees
• FEATURE_NUMBER: feature number of data
• FEATURE_NUM_FOR_TRAIN: number of randomly selected features for each bagging
• TRAINING_SET_SIZE: number of items in training set
• TRAINING_SET_SIZE_FOR_TRAIN: number of randomly selected training items for each bagging
• TESTING_SET_SIZE: number of items in testing set
• CROSS_VALIDATION_SIZE: number of training set items used to perform cross validation
• SCATTER_RATIO: ratio of scatter number to size of data, used to generate random subset of training items and features

Openmp Parallelization

This program use Openmp to enable parallelization in:

• finding feature to split nodes
• generating result of testing set

To disable parallelization, modify USE_OPENMP := 1 to USE_OPENMP := 0 in Makfile, and remake.

Run the Program

• cd src
• make
• ./main.out