Kernel Density Estimation
We perform density estimations with Gaussian kernels for each individual word, allowing calculation of the probability of being assigned to a network class given a particular value of the considered word. By comparing ratios of likelihood values among the different models, it is therefore possible to determine by which network model the feature value of the real network is most likely generated.
The following shows density estimations for the words having maximal and minimal likelihood ratio between the winner (by SVM) model Middendorf-Ziv (MZ) (duplication-mutation) and the runner-up model Grindrod (static).
nnz D(AUTAUTAUA), (associated subgraphs), maximum likelihood ratio
nnz D(AUADATAUA), (associated subgraphs), minimum likelihood ratio
The following shows density estimations for the words having maximal and minimal likelihood ratio between the winner (by SVM) model Kumar (duplication-mutation) and the runner-up model Krapivsky-Bianconi (preferential attachment).
nnz D(AUTAUTAUAUTA), (associated subgraphs), maximum likelihood ratio
nnz U(AUTA), (associated subgraphs), minimum likelihood ratio
The following shows density estimations for the words having maximal and minimal likelihood ratio between the winner (by SVM) model Sole (duplication-mutation) and the runner-up model Vazquez (duplication-mutation).
nnz U(AADAAA), (associated subgraphs), maximum likelihood ratio
nnz D(AAA), (associated subgraphs), minimum likelihood ratio
