sivi_lr: SIVI for Logistic Regression

In wanghg1996/SIVIR: SEMI-IMPLICIT VARIATIONAL INFERENCE

Description

This function is used for the implementation of the SIVI's algorithm (semi-implicit variational inference) in the logistic regression set-up. It adopts the adam optimizer for the covariance matrix and uses the gradient descent optimizer for the Neural Network in the implicit part. The iter number can be tuned manually, and the output is the sampling from the posterior of the latent variables. In this configuration, we have the number of latent variables equal to the numbers of parameters input by X.

Usage

sivi_lr(X, Y, noise_dim = 20, K = 50, J = 100, merge = 1,
  lr = 1e-04, lr2 = 0.01, n_iter = 500, inner_iter = 200,
  pos_num = 1000)

Arguments

X	the matrix for training
Y	the corresponding labels
noise_dim	the input noise dimension
K	the added terms
J	the sampling numbers of latent variables
merge	when merge equals one means using the lower bound, when it equals zero means using the upper bound
lr	the learning rate inside the 200 times iteration
lr2	the learning rate outside the 200 times iteration
n_iter	the whole iteration number
inner_iter	the iteration number for the nural network which reside in the inner part of the iteration and is supposed to be smaller than the iter_num.
pos_num	the sampling numbers of the posterior

Value

the sampling of posterior for the latent variables

Examples

sivi_lr(waveform$X.train, waveform$y.train)

wanghg1996/SIVIR documentation built on Dec. 3, 2019, 7:57 a.m.