bobye / decision_tree_pdeque

C++ code for "A Faster Drop-in Implementation for Leaf-wise Exact Greedy Induction of Decision Tree Using Pre-sorted Deque"

Leaf-wise Induction of Decision Tree with Presorted Deque

This is the proof-of-concept demo code for reproducing experiments in the arXiv note "A Faster Drop-in Implementation for Leaf-wise Exact Greedy Induction of Decision Tree Using Pre-sorted Deque" (https://arxiv.org/abs/1712.06989).

Prepare sample data

• download data from HIGGS and uncompress gz file.
• create training data head -1000000 HIGGS.csv > higgs-train-1m.csv
• create testing data tail -50000 HIGGS.csv > higgs-test.csv

Compile and test

$ make
$ OMP_NUM_THREADS=28 ./build/test_dt higgs-train-1m.csv higgs-test.csv
tree induction time: 1.475672 seconds
training time: 2.048821 seconds
nleafs: 1845 
test metric: FP 0.276, FN 0.317, Sensitivity 0.720, Specificity 0.687, Accuracy 0.705

Other tests on synthetic data

$ OMP_NUM_THREADS=12 ./build/test_dt 
tree induction time: 1.425927 seconds
training time: 2.048105 seconds
nleafs: 24 
test metric: FP 0.801, FN 0.000, Sensitivity 1.000, Specificity 0.985, Accuracy 0.985

---

All rights reserved (2017-2023). Jianbo Ye

decision_tree_pdeque's Issues

implement random forest

build the first template library for decision trees

code refactor

documentation

enable customizable cost function for split cut

model serialization, communication, and visualization

implement tree pruning

implement priority tree growth

extend codes for building regression tree