A touch on the subject of likelihood
-
Need to take advantage of simulations (model and parameter sets) that did well in comparison to the real world.
-
In fact up to this point we have not made any statements about how this relates to the "God" model.
-
It would be beneficial to weight the simulations results by how well they matched measurements before a distribution is produced.
-
The likelyhood estimator of doff=1/1+sqrt((distpredicted-distgod)2) is used.
-
Required only that it be monitonically increasing as the simulation is closer to the actual answer.
<< >>