Discontinuous design space

lukav · 19 May 2021 21:34

Hi,

I am working on an optimisation in a highly non-linear design space.
The idea is to create a surrogate model with least samples possible. I have 200 samples for now, there are 16 design parameters and 1 quantity of interest.
What would you recommend me to try from your code?

Thanks,

BTW I put my DOE table here in the case you want to check it out.

github.com

lukavincekovic/mim_dataset/blob/c3e772cc614929a51eec53d2500ae0315563f80e/dataset_200.csv

,v1,v2,v3,v4,v5,v6,v7,v8,v9,v10,v11,v12,v13,v14,v15,v16,qoi1
0,-0.2830189,-0.2704403,0.4842767,-0.509434,-0.3710692,0.5345912,0.7358491,0.4465409,0.1823899,0.5849057,-0.4842767,0.9371069,0.4716981,0.4213836,0.5345912,-0.3207547,9.93972
1,-0.4968553,-0.9874214,-0.3836478,-0.6855346,0.6100629,0.6603774,-0.9245283,-0.5345912,-0.4842767,0.2327044,-0.5597484,-0.3584906,0.1572327,-0.9622642,-0.03144654,-0.4213836,11.44351
2,-0.6100629,-0.08176101,0.01886792,-0.9496855,0.5220126,-0.8742138,0.9874214,-0.7610063,-0.8238994,-0.7987421,0.1823899,-0.7735849,-0.3584906,0.7358491,-0.5471698,0.5597484,9.34903
3,0.1949686,-0.9622642,-0.1572327,0.9119497,-0.3459119,0.9496855,0.5345912,-0.4339623,-0.3081761,0.3081761,-0.8490566,0.7232704,0.1823899,0.5849057,-0.8616352,-0.509434,9.25438
4,1,-0.6603774,0.836478,-0.9245283,0.01886792,0.1823899,0.5471698,-0.5597484,-0.3207547,0.6100629,-0.1069182,0.6226415,0.8616352,-0.672956,-0.2075472,0.5345912,9.6591
5,-0.7735849,-0.3459119,-0.2955975,0.8993711,-0.6855346,-0.4716981,-0.4716981,-0.8490566,0.3081761,0.3584906,0.006289308,-0.2830189,-0.1194969,-0.9371069,0.9119497,0.9622642,9.10686
6,0.836478,0.1069182,0.4465409,-0.8993711,-0.3584906,0.3584906,0.8867925,0.8993711,-0.1069182,-0.7610063,-0.408805,-0.3333333,-0.6477987,-0.6226415,-0.4213836,0.4465409,9.1348
7,0.8742138,0.4842767,-1,0.7987421,-0.8238994,-0.1069182,0.8238994,0.3710692,-0.5345912,-0.01886792,-0.2830189,-0.1069182,0.5974843,-0.8616352,-0.8867925,0.6603774,9.385
8,0.2201258,-0.09433962,-0.1069182,-0.8616352,0.4968553,-0.7106918,-0.8867925,0.06918239,0.4465409,0.3459119,0.1320755,-0.1823899,-0.8616352,-0.08176101,-0.7735849,-0.7610063,9.5753
9,-0.6603774,-0.1698113,-0.3584906,-0.8490566,0.5471698,-0.1194969,-0.7610063,-0.4968553,-0.9119497,0.006289308,0.9245283,-0.2578616,-0.05660377,0.6100629,-0.4842767,-0.8113208,9.0659
10,-0.3836478,0.3710692,-0.8238994,-0.2327044,0.2201258,0.3836478,0.4968553,-0.3836478,-0.7106918,-0.1069182,-0.01886792,0.2075472,-0.8742138,0.1698113,-0.6352201,0.7610063,8.9826
11,0.4591195,0.1823899,-0.5471698,-0.5974843,-0.04402516,-0.6352201,0.09433962,-0.3081761,0.2201258,-0.7484277,-0.9119497,-0.9874214,0.509434,0.9245283,-0.5849057,-0.4591195,9.98746
12,-0.7106918,-0.3836478,0.3333333,0.4213836,0.2955975,-0.4213836,0.4213836,-0.6603774,0.245283,0.8867925,0.3081761,-0.5220126,0.6855346,-0.1949686,0.1572327,-0.06918239,10.16116
13,0.509434,0.2578616,-0.2704403,0.4591195,0.6352201,0.2955975,0.5597484,0.7358491,-0.3459119,0.5220126,-0.2955975,0.09433962,0.3333333,-0.7735849,-0.1823899,0.1823899,9.0321
14,0.5345912,-0.7610063,0.9874214,0.2075472,0.9622642,0.8238994,-0.2578616,-0.9119497,-0.4716981,0.9496855,0.3333333,-0.3836478,0.9496855,-0.2201258,-0.6100629,-0.04402516,9.51124
15,-0.7610063,0.006289308,-0.1320755,0.04402516,0.3962264,0.8616352,-0.509434,0.8742138,-0.8993711,-0.05660377,-0.7610063,0.1194969,1,-0.3333333,-0.9496855,-0.6855346,11.3519
16,0.8238994,-0.6477987,0.3836478,0.8616352,-0.4213836,0.4339623,-0.2955975,0.5849057,0.3962264,0.3207547,-0.9496855,-0.7987421,0.2704403,0.408805,0.3836478,-0.3836478,9.5696
17,0.8113208,0.509434,-0.5974843,-0.4968553,-0.2201258,0.4213836,-0.672956,0.7484277,-0.9245283,0.3710692,0.4842767,0.7610063,0.01886792,-0.509434,0.8616352,0.6352201,9.6996
18,0.9371069,0.3459119,-0.4339623,-0.9622642,0.8993711,-0.7610063,-0.7735849,0.3333333,0.9622642,-0.9748428,0.1949686,0.509434,-0.8238994,-0.836478,0.5849057,-0.3459119,9.05789

This file has been truncated. show original

psesh · 21 May 2021 08:36

Hi @lukav, have you tried fitting a standard least squares / compressed sensing / elastic-net polynomial to your data? You can start with order one and then perhaps try order two. Cheers!