Skip to content

parameters.md

Latest commit

 

History

History
82 lines (46 loc) · 3.21 KB

parameters.md

File metadata and controls

82 lines (46 loc) · 3.21 KB

ThunderGBM Parameters

This page is for parameter specification in ThunderGBM. The parameters used in ThunderGBM are identical to XGBoost (except some newly introduced parameters), so existing XGBoost users can easily get used to ThunderGBM.

Key arameters for both python and c++ command line

  • verbose [default=1]

    • Printing information: 0 for silence, 1 for key information and 2 for more information.

  • depth [default=6]

    • The maximum depth of the decision trees. Shallow trees tend to have better generality, and deep trees are more likely to overfit the training data.

  • n_trees [default=40]

    • The number of training iterations. n_trees equals to the number of trees in GBDTs.

  • n_gpus [default=1]

    • The number of GPUs to be used in the training.

  • max_num_bin [default=255]

    • The maximum number of bins in a histogram.

  • column_sampling_rate [default=1]

    • The sampling ratio of subsampling columns (i.e., features)

  • bagging [default=0]

    • This option is for training random forests. Setting it to 1 to perform bagging.

  • n_parallel_trees [default=1]

    • This option is used for random forests to specify how many trees per iteration.

  • learning_rate [default=1, alias(only for c++): eta]

    • valid domain: [0,1]. This option is to set the weight of newly trained tree. Use eta < 1 to mitigate overfitting.

  • objective [default="reg:linear"]

    • valid options include reg:linear, reg:logistic, multi:softprob, multi:softmax, rank:pairwise and rank:ndcg.
    • reg:linear is for regression, reg:logistic and binary:logistic are for binary classification.
    • multi:softprob and multi:softmax are for multi-class classification. multi:softprob outputs probability for each class, and multi:softmax outputs the label only.
    • rank:pairwise and rank:ndcg are for ranking problems.

  • num_class [default=1]

    • set the number of classes in the multi-class classification. This option is not compulsory.

  • min_child_weight [default=1]

    • The minimum sum of instance weight (measured by the second order derivative) needed in a child node.

  • lambda_tgbm [default=1, alias(only for c++): lambda or reg_lambda]

    • L2 regularization term on weights.

  • gamma [default=1, alias(only for c++): min_split_loss]

    • The minimum loss reduction required to make a further split on a leaf node of the tree. gamma is used in the pruning stage.

  • tree_method [default="auto"]

    • "auto": select the approach of finding best splits using the builtin heuristics.

    • "exact": find the best split using enumeration on all the possible feature values.

    • "hist": find the best split using histogram based approach.

Parameters only for c++ command line:

  • data [default="../dataset/test_dataset.txt"]

    • The path to the training data set

  • model_out [default="tgbm.model"]

    • The file name of the output model. This option is used in training.

  • model_in [default="tgbm.model"]

    • The file name of the input model. This option is used in prediction.