You are on page 1of 5

! README.

md

Simulations Regarding Self-Subtraction


These simulations are testing how MLE works when we know for a fact there is no self-subtraction in the design vectors. To ensure
this, the strategy is to get the design vectors from the base rather than the injected cube .

Expectations
I expect a few things- most importantly, I expect there to be differences in how the MLE estimates change as the number of
components change. Theoretically, the fit amplitude should be completely independent since we can't self-subtract any flux.

Results

Parameter Estimates
Grid

Filtered Grid
Sigmas

Filtered Sigmas
Model Selection
Evidence

Filtered Evidence
Note: In this setup, the BIC and AIC only differ by a constant-

BIC = k * ln(N) - 2 * ln(L)


AIC = 2k - 2 * ln(L)
BIC - AIC = k * ln(N) - 2k

BIC/AIC

Filtered BIC/AIC

BIC/AIC vs No Companion
Note: In this case, the BIC and AIC for the no-companion model are the same, since k=0 .

You might also like