- LightGBM.ipynb: Modified version of marugari's work
- exp013
- model : XGB(hist_depthwise, hist_lossguie, hist_GPU, GPU), LGB
- objective : Binary classification
- metric : Logloss
- dataset : make_classification
- n_train : 0.5M, 1M, 2M, 4M
- n_valid : n_train/4
- n_features : 32
- n_clusters_per_class : 8
- n_rounds : 100
- max_depth : 5, 10, 15
- num_leaves : 2 ** max_depth
- exp014
- model : XGB(hist_depthwise, hist_lossguie, hist_GPU, GPU), LGB
- objective : Binary classification
- metric : Logloss
- dataset : make_classification
- n_train : 1,2,4,8,16,32,64 * 10K
- n_valid : n_train/4
- n_features : 256
- n_clusters_per_class : 8
- n_rounds : 100
- max_depth : 5, 10
- num_leaves : 2 ** max_depth
The following codes were run on older versions of XGBoost and LightGBM
- exp010
- model : XGB(CPU, EQBIN_depthwise, EQBIN_lossguie, GPU), LGB
- objective : Binary classification
- metric : Logloss
- dataset : make_classification
- n_train : 0.5M, 1M, 2M
- n_valid : n_train/4
- n_features : 32
- n_rounds : 100
- n_clusters_per_class : 8
- max_depth : 5, 10, 15
- exp011
- model : XGB(EQBIN_depthwise, EQBIN_lossguie), LGB
- objective : Binary classification
- metric : Logloss
- dataset : make_classification
- n_train : 0.5M, 1M, 2M
- n_valid : n_train/4
- n_features : 32
- n_clusters_per_class : 8
- n_rounds : 200
- max_depth : 5, 10, 15, 20
- num_leaves : 32, 256, 1024, 4096, 16384
- exp012
- model : XGB(EQBIN_depthwise, EQBIN_lossguie), LGB
- objective : Binary classification
- metric : Logloss
- dataset : make_classification
- n_train : 1, 2, 4, 8, 16, 32 * 10000
- n_valid : n_train/4
- n_features : 256
- n_clusters_per_class : 8
- n_rounds : 100
- max_depth : 5, 10, 15, 20
- num_leaves : 32, 256, 1024, 4096