|
|
Training data worsens fit of mixture ... |
|
Message/Author |
|
|
i have fit several sets of models with and without training data. in each case, introducing the training data has worsened model fit as measured by BIC. should i conclude from this that the training data is not very accurate (a possibility) or is there a statistical reason for this phenomenon? |
|
|
The metric of the likelihood changes when training data are introduced, so BIC values are not comparable with and without training data. |
|
|
I assume the same would hold for two different sets of training data, i.e., the BIC would not be comparable. |
|
|
Right. |
|
Back to top |
|
|