File "noise"

 This file provides a random forest method to determine a latent space, an encoder and a decoder. The method uses branch and bound where the failure count of a bayes classifier learnt decoder is used as the loss function. The failure count is also used to randomly blame a single path of the encoder. And then perturb it by a Bernoulli trail.

The following noise predicates are provided:

noise_train(E, Z, N, I, T, B):
The predicate succeeds in B with the binary decision diagram and in T with the failure score, for the transformer encoder of the mapping P-Q from data(Z,P,Q). The parameter N gives the ensemble size, the parameter E the initial encoder and I the number of iterations.
noise_recon(N, E, Z, S, C):
The predicate succeeds in C with the binary decision diagram and in S with the failure score, for the transformer decoder of the encoder E and the mapping P-Q from data(Z,P,Q). The parameter N gives the ensemble size.

Kommentare