Return a 4D shape Args: in_val (int or list with length 2) Returns: list with length 4

Function for load training data If n_use_label or n_use_sample is not None, samples will be randomly picked to have a balanced number of ex

modify from: https://github.com/fastforwardlabs/vae-tf/blob/master/plot.py#L45

create samping model in fig 1 and 3 in the paper

create training model in fig 1, 3 and 6 in the paper

Return a 2D shape Args: in_val (int or list with length 2) Returns: list with length 2

Function for load validation data

create input for sampling model in fig 1, 3 and 6 in the paper

create input for training model in fig 1, 3, 6 and 8 in the paper

estimate digit label for semi-supervised model

create samping model in fig 6 in the paper

create input for testing model in fig 8 in the paper

Flatten the tensor except the first dimension.

create testing model in fig 8 in the paper

create training model in fig 8 in the paper

function for sampling images from trained model

This function is copied from `tensorpack <https://github.com/ppwwyyxx/tensorpack/blob/master/tensorpack/utils/utils.py>`__. Get a good RN

Function for semisupervised training (Fig 8 in the paper) Validation will be processed after each epoch of training Loss of each modules wi

Function for supervised training (Fig 6 in the paper) Validation will be processed after each epoch of training. Loss of each modules will b

Function for unsupervised training and incorporate label info in adversarial regularization (Fig 1 and 3 in the paper) Validati

function for visualize latent space of trained model when ncode = 2

Args: im_size (int or list of length 2): size of input image n_channel (int): number of input image chann