Metrics¶
-
class
libspn_keras.metrics.
LogLikelihood
(name='llh', **kwargs)¶ Compute log marginal \(1/N \sum \log(p(X))\).
Assumes that the last layer of the SPN is a
RootSum
. It ignores they_true
argument, since a target for \(Y\) is absent in unsupervised learning.