layer-class: Layer
In davharris/mistnet: Stochastic Neural Networks

A reference class object for one layer of computation in a network object.

coef.dim: a length-two integer vector
weights: a matrix of real numbers
biases: a numeric vector containing the intercept for each node
nonlinearity: a nonlinearity object
prior: a prior object
inputs: a numeric array with the input activity to each node in response to each example, for each Monte Carlo sample
outputs: a numeric array with the transformed activations for each node in response to each example, for each Monte Carlo sample
error.grads: a numeric array
weighted.bias.grads: a numeric vector
weighted.llik.grads: a numeric matrix
coef.updater: an updater object

backwardPass(incoming.error.grad, sample.num): Calculate error.grads for one sample
combineSampleGrads(inputs, weights, n.importance.samples): update weighted.llik.grads and weighted.bias.grads based on importance weights and gradients from backpropagation
forwardPass(input, sample.num): Update inputs and outputs for one sample
resetState(n.minibatch, n.importance.samples): Reset inputs, outputs, and error.grads to NA; alter the minibatch size and number of importance samples if desired
updateCoefficients(dataset.size, n.minibatch): Calculate coef.delta and add it to weights. Update biases