Skip to content

Commit 3b21468

Browse files
ahuber21ahuber21
authored
Add SHAP value benchmarks (#144)
* Add SHAP value benchmarks * Add report generator config for MB reports * black format * review comments * Remove nthreads -1 logic from xgb_mb runner
1 parent f146815 commit 3b21468

File tree

3 files changed

+376
-14
lines changed

3 files changed

+376
-14
lines changed

‎configs/modelbuilders/xgb_mb_cpu_config_shap.json

Lines changed: 309 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,309 @@
1+
{
2+
"common": {
3+
"lib": "modelbuilders",
4+
"data-format": "pandas",
5+
"data-order": "F",
6+
"dtype": "float32",
7+
"algorithm": "xgb_mb",
8+
"tree-method": "hist",
9+
"count-dmatrix": "",
10+
"num-threads": -1,
11+
"n-estimators": 50
12+
},
13+
"cases": [
14+
{
15+
"dataset": [
16+
{
17+
"source": "npy",
18+
"name": "abalone",
19+
"training": {
20+
"x": "data/abalone_x_train.npy",
21+
"y": "data/abalone_y_train.npy"
22+
},
23+
"testing": {
24+
"x": "data/abalone_x_test.npy",
25+
"y": "data/abalone_y_test.npy"
26+
}
27+
}
28+
],
29+
"learning-rate": 0.03,
30+
"max-depth": 6,
31+
"n-estimators": 1000,
32+
"objective": "reg:squarederror"
33+
},
34+
{
35+
"dataset": [
36+
{
37+
"source": "npy",
38+
"name": "mortgage1Q",
39+
"training": {
40+
"x": "data/mortgage1Q_x_train.npy",
41+
"y": "data/mortgage1Q_y_train.npy"
42+
}
43+
}
44+
],
45+
"n-estimators": 100,
46+
"objective": "reg:squarederror",
47+
"max-depth": 8,
48+
"scale-pos-weight": 2,
49+
"learning-rate": 0.1,
50+
"subsample": 1,
51+
"reg-alpha": 0.9,
52+
"reg-lambda": 1,
53+
"min-child-weight": 0,
54+
"max-leaves": 256
55+
},
56+
{
57+
"objective": "reg:squarederror",
58+
"max-depth": 8,
59+
"learning-rate": 0.1,
60+
"reg-lambda": 1,
61+
"max-leaves": 256,
62+
"dataset": [
63+
{
64+
"source": "npy",
65+
"name": "year_prediction_msd",
66+
"training": {
67+
"x": "data/year_prediction_msd_x_train.npy",
68+
"y": "data/year_prediction_msd_y_train.npy"
69+
},
70+
"testing": {
71+
"x": "data/year_prediction_msd_x_test.npy",
72+
"y": "data/year_prediction_msd_y_test.npy"
73+
}
74+
}
75+
]
76+
},
77+
{
78+
"objective": "reg:squarederror",
79+
"max-depth": 6,
80+
"learning-rate": 0.1,
81+
"reg-lambda": 1,
82+
"max-leaves": 1024,
83+
"dataset": [
84+
{
85+
"source": "synthetic",
86+
"type": "regression",
87+
"n_features": 10,
88+
"n_informative": 10,
89+
"training": {
90+
"n_samples": 2000
91+
},
92+
"testing": {
93+
"n_samples": 8000
94+
}
95+
}
96+
]
97+
},
98+
{
99+
"objective": "reg:squarederror",
100+
"max-depth": 10,
101+
"learning-rate": 0.1,
102+
"reg-lambda": 1,
103+
"max-leaves": 4096,
104+
"dataset": [
105+
{
106+
"source": "synthetic",
107+
"type": "regression",
108+
"n_features": 10,
109+
"n_informative": 10,
110+
"training": {
111+
"n_samples": 3000
112+
},
113+
"testing": {
114+
"n_samples": 80000
115+
}
116+
}
117+
]
118+
},
119+
{
120+
"objective": "reg:squarederror",
121+
"max-depth": 8,
122+
"learning-rate": 0.1,
123+
"reg-lambda": 1,
124+
"max-leaves": 256,
125+
"dataset": [
126+
{
127+
"source": "synthetic",
128+
"type": "regression",
129+
"n_features": 20,
130+
"n_informative": 20,
131+
"training": {
132+
"n_samples": 2000
133+
},
134+
"testing": {
135+
"n_samples": 80000
136+
}
137+
}
138+
]
139+
},
140+
{
141+
"objective": "reg:squarederror",
142+
"max-depth": 10,
143+
"learning-rate": 0.1,
144+
"reg-lambda": 1,
145+
"max-leaves": 1024,
146+
"dataset": [
147+
{
148+
"source": "synthetic",
149+
"type": "regression",
150+
"n_features": 20,
151+
"n_informative": 20,
152+
"training": {
153+
"n_samples": 4000
154+
},
155+
"testing": {
156+
"n_samples": 80000
157+
}
158+
}
159+
]
160+
},
161+
{
162+
"objective": "reg:squarederror",
163+
"max-depth": 14,
164+
"learning-rate": 0.1,
165+
"reg-lambda": 1,
166+
"max-leaves": 4096,
167+
"dataset": [
168+
{
169+
"source": "synthetic",
170+
"type": "regression",
171+
"n_features": 20,
172+
"n_informative": 20,
173+
"training": {
174+
"n_samples": 10000
175+
},
176+
"testing": {
177+
"n_samples": 80000
178+
}
179+
}
180+
]
181+
},
182+
{
183+
"objective": "reg:squarederror",
184+
"max-depth": 8,
185+
"learning-rate": 0.1,
186+
"reg-lambda": 1,
187+
"max-leaves": 256,
188+
"dataset": [
189+
{
190+
"source": "synthetic",
191+
"type": "regression",
192+
"n_features": 50,
193+
"n_informative": 50,
194+
"training": {
195+
"n_samples": 2000
196+
},
197+
"testing": {
198+
"n_samples": 80000
199+
}
200+
}
201+
]
202+
},
203+
{
204+
"objective": "reg:squarederror",
205+
"max-depth": 10,
206+
"learning-rate": 0.1,
207+
"reg-lambda": 1,
208+
"max-leaves": 1024,
209+
"dataset": [
210+
{
211+
"source": "synthetic",
212+
"type": "regression",
213+
"n_features": 50,
214+
"n_informative": 50,
215+
"training": {
216+
"n_samples": 2000
217+
},
218+
"testing": {
219+
"n_samples": 80000
220+
}
221+
}
222+
]
223+
},
224+
{
225+
"objective": "reg:squarederror",
226+
"max-depth": 14,
227+
"learning-rate": 0.1,
228+
"reg-lambda": 1,
229+
"max-leaves": 4096,
230+
"dataset": [
231+
{
232+
"source": "synthetic",
233+
"type": "regression",
234+
"n_features": 50,
235+
"n_informative": 50,
236+
"training": {
237+
"n_samples": 4000
238+
},
239+
"testing": {
240+
"n_samples": 80000
241+
}
242+
}
243+
]
244+
},
245+
{
246+
"objective": "reg:squarederror",
247+
"max-depth": 8,
248+
"learning-rate": 0.1,
249+
"reg-lambda": 1,
250+
"max-leaves": 256,
251+
"dataset": [
252+
{
253+
"source": "synthetic",
254+
"type": "regression",
255+
"n_features": 100,
256+
"n_informative": 100,
257+
"training": {
258+
"n_samples": 1000
259+
},
260+
"testing": {
261+
"n_samples": 80000
262+
}
263+
}
264+
]
265+
},
266+
{
267+
"objective": "reg:squarederror",
268+
"max-depth": 10,
269+
"learning-rate": 0.1,
270+
"reg-lambda": 1,
271+
"max-leaves": 1024,
272+
"dataset": [
273+
{
274+
"source": "synthetic",
275+
"type": "regression",
276+
"n_features": 100,
277+
"n_informative": 100,
278+
"training": {
279+
"n_samples": 2000
280+
},
281+
"testing": {
282+
"n_samples": 80000
283+
}
284+
}
285+
]
286+
},
287+
{
288+
"objective": "reg:squarederror",
289+
"max-depth": 14,
290+
"learning-rate": 0.1,
291+
"reg-lambda": 1,
292+
"max-leaves": 4096,
293+
"dataset": [
294+
{
295+
"source": "synthetic",
296+
"type": "regression",
297+
"n_features": 100,
298+
"n_informative": 100,
299+
"training": {
300+
"n_samples": 3000
301+
},
302+
"testing": {
303+
"n_samples": 80000
304+
}
305+
}
306+
]
307+
}
308+
]
309+
}

0 commit comments

Comments
 (0)