TABLE 1.
Hyperparameter settings of LightGBM used for the Bayesian optimization.
Parameters | Search space |
---|---|
learning_rate | (1e-8, 1.0,'log-uniform') |
lambda_l1 | (1e-8, 1.0,'log-uniform') |
lambda_l2 | (1e-6, 1.0,'log-uniform') |
bagging_fraction | (0.3, 1.0,'uniform') |
feature_fraction | (0.3, 1.0,'uniform') |
bagging_freq | (2, 10,'int') |
min_child_samples | (1, 50,'int') |
max_depth | (2, 20,'int') |
num_leaves | (2, 1024,'int’, step = 2) |
n_estimators | (100, 4000,'int’, step = 5) |