Description Usage Arguments Value
Implementation based on http://web.stanford.edu/class/archive/cs/cs294a/cs294a.1104/sparseAutoencoder.pdf
1 2 3 4 |
X |
matrix. Training dataset. |
num_hidden |
integer. Specifies number of neurons in each hidden layer. |
activation |
character or function. If it's a character it has to be one of the
predefined activation functions. If it's a function, |
lambda |
numeric. Weight decay parameter. |
beta |
numeric. Learning rate parameter for algorithm trying to (approximately) satisfy the sparsity constraint. |
rho |
numeric. Sparsity parameter, which specifies our desired level of sparsity. |
epsilon |
numeric. A (small) parameter for initialization of weight matrices
as small gaussian random numbers sampled from |
tolerance |
numeric (optional). Tolarance to be used for comparing floting point numbers.
Default is |
X.test |
matrix (optional). Testing dataset for evaluating the network. |
d_activation |
function (optional). This parameter will be omitted if activation is a character, otherwise this function will be used in backpropagation as activation prime, in which case this parameter becomes mandatory. |
optim_method |
character (optional). Optimization method to be used. Please check
|
max_iterations |
integer (optional). Maximum number of iterations for
optimizer. Default is |
rescale |
logical or numeric. Autoencoders yield better results when used on
normalized matrices. Normalization should be performed according to the activation
function being used. Default is |
rescaling_offest |
numeric. A small value used in rescaling to
|
Object of class autoenc.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.