Closed Erika1kuta closed 10 months ago
Sorry for the late response, but I'm no longer in academia and can't maintain this repo very actively. But I think to answer your questions we have to consider two aspects:
In fact, the optimum is close to zero, as the cost surface suggests: So, this part of the result seems correct.
The negative result is caused by the fact that the GMM defines the model of the measurement's probability.
And since the cost function is defined as f(x) = x - z
, a negative x is the optimal solution for a positive mean in z.
But I understand that you expect that the GMM describes the distribution of x
in this example.
It is a bit counterintuitive, but usually we want to describe the measurement and not the state with our mixture model.
Hopefully this answers your question – feel free to reopen if required.
Best Regards Tim
Dear Tim, Thank you very much for your contribution. Perhaps I didn't quite understand when I followed your example:
./App_Robust_Models_2D empty empty Data_2D_Output.txt 10 8 MaxSumMix 0 0 1 2 0.5 0 0 1 2 0 0 5 0.35 0.65
I have obtained the following data:My confusion is why all optimization points are negative and near (0,0), because according to the weighted ratio of 0.35 and 0.65 and covariance, (0,0) and (1,2) their Gaussian mixture models should not be within this range. Perhaps I didn't quite understand what you meant. Looking forward to your reply. Many thanks.