Namespace: Models

A layer of neurons (perceptrons).

Nested types and modules


Neural layer hyper-parameters.


Neural layer parameters.

Functions and values

Function or valueDescription
Signature: HyperPars
pars mb hp
Signature: mb:ModelBuilder<'?176100> -> hp:HyperPars -> Pars<'?176101>
Type parameters: '?176100, '?176101

Creates the parameters for the neural-layer in the supplied model builder mb using the hyper-parameters hp. The weights are initialized using random numbers from a uniform distribution with support [-r, r] where r = 4 * sqrt (6 / (hp.NInput + hp.NOutput)). The biases are initialized to zero.

pred pars input
Signature: pars:Pars<'?176103> -> input:ExprT -> ExprT
Type parameters: '?176103

Returns an expression for the output (predictions) of the neural layer with parameters pars given the input input. If the soft-max transfer function is used, the normalization is performed over axis 0.

regularizationTerm pars
Signature: pars:Pars<'?176105> -> ExprT
Type parameters: '?176105

Calculates sum of all regularization terms of this layer.

Fork me on GitHub