model configs:

Regression:

For deep ensemble

100 samples, 50 models, 400 epochs: 153 (First Case)

100 samples, 10 models, 400 epochs: 32 (Second Case)

100 samples, 5 models, 400 epochs: 13 (Third Case)

Equivalents (equivalents must use same number of training samples, the rest can be different):

First Case

Bootstrap 100, 50, 1000 (165)

MC Dropout 100, 5/10/50, 1000 (7) is the longest it takes to train

LA 100, 100/250/1000, 1000 (5) is the longest it takes to train

HMC 100, 1000, 1000 (212)

VI 100, 1000, 1000 (22) is the longest it takes to train

Second Case

same for all except for HMC and Bootstrap

HMC 100, 100, 400 (32)

Bootstrap 100, 50, 200 (37)

Third Case

same for all except for HMC and Bootstrap

HMC 100, 100, 1000 (24)

Bootstrap 100, 5, 400 (16)

For deep ensemble

30 samples, 50 models, 400 epochs: 36 (First Case)

30 samples, 10 models, 400 epochs: 7 (Second Case)

30 samples, 5 models, 400 epochs: 4 (Third Case)

Equivalents (equivalents must use same number of training samples, the rest can be different):

First Case

Bootstrap 30, 50, 400 (46)

MC Dropout 30, 5/10/50, 1000 (3) is the longest it takes to train

LA 30, 100/250/1000, 1000 (2) is the longest it takes to train

HMC 30, 100, 400 (27)

Second Case

same for all except for HMC and Bootstrap

HMC 30, 100, 200 (24)

Bootstrap 30, 10, 400 (7)

Third Case

same for all except for HMC and Bootstrap

HMC 30, 100, 200 (24)

Bootstrap 30, 5, 400 (4)