Rejection rates in DEzs about bayesiantools

[ForModLabUHel]

To remove the unreasonable points we could also try to implement in DE the algorithm that is used in DREAM to remove the outliers and limit its use in the burn-in phase to avoid problems in a detailed balance proof.

[florianhartig]

Yes, that is another option.

In any case, we could also use #8 during adaptation only

[stefan-paul]

We already have the sampler @checcomi described.
If you set nCR = 1 and DEpairs = 2 in DREAM then it is
basically a DE sampler that removes outliers during burn-in.

[florianhartig]

good point. I will try this!

[ForModLabUHel]

@stefan-paul good!
Does the outlier remover works only in the burn-in?

[stefan-paul]

@checcomi yes.

[florianhartig]

I'll close this now, this is too diffuse ... most of this is probably addressed by the adaptation issue #87

As for the rest - if DEzs has low acceptance, it's probably an issue of the settings or of the sampler in general, for the particular function that it is used on.