NLL: Negative Log Likelyhood

In frhl/neuralnetr: Neural Nets in R

Description

Each loss module has a forward method that takes in a batch of predictions Ypred (from the previous layer) and labels Y and returns a scalar loss value.

The NLL module has a backward method that returns dLdZ, the gradient with respect to the preactivation to SoftMax (note: not the activation!), since we are always pairing SoftMax activation with NLL loss

Super class

neuralnetr::ClassModule -> NLL

Public fields

Ypred

the predicted Y vector

Y

the actual Y vector

Methods

Public methods

• NLL$forward()

• NLL$backward()

• NLL$clone()

Inherited methods

• neuralnetr::ClassModule$sgd_step()

---

Method forward()

Calculate the loss.

Usage

NLL$forward(Ypred, Y)

Arguments

Ypred

the predicted Y vector.

Y

the actual Y vector.

Returns

loss as a scalar

---

Method backward()

calculate the gradient w.r.t. the loss

Usage

NLL$backward()

Returns

dLdZ (? x b)

---

Method clone()

The objects of this class are cloneable with this method.

Usage

NLL$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

See Also

Other loss: SquaredLoss

frhl/neuralnetr documentation built on Nov. 9, 2020, 2:24 p.m.