Decision Tree Train enviroment context struct.
Random Forest Train enviroment context struct.
Note that the structure are used in both host and kernel
As shown in the struct, there are 8 memebers.
- cretiea is the algorithm to compute node impurity and information gain 0 : gini impurity(support) 1 : entropy (not support) 2 : Variance (not support)
- min_leaf_size A node stops splitting when it receives the number of training instances
- max_tree_depth is user input value in host, not transfer to device.
- (for extension) maxBins determines the max splits of each feature. in current version, splits are read from configuration. The parameter is for splits generation extension
- (for extension) min_samplecount_for_sample, for random forest and split generation sampling
- (for extension) sample_percent_if_sample, for random forest and split generation sampling
- (for extension) min_info_gain For entropy extension, a node stops splitting when its max gain <= this value
// fields unsigned cretiea unsigned max_tree_depth unsigned min_leaf_size float max_leaf_cat_per float min_info_gain int min_samplecount_for_sample float sample_percent_if_sample unsigned maxBins float fea_fraction ap_uint <64> seed