Annex v0.2.1 Annex.Cost.MeanSquaredError View Source
MeanSquaredError is the module that encapsulates the calculation of the mean squared error equation.
The mean squared error is a calculation that describes the penalty (Cost
)
of the distance between points that highly penalizes large values and
slightly penalizes small values by applying the square (x * x
) of the
difference between the expected values (labels
) and the predicted values
(predictions
) of a sequence/network/model, summing the squares and taking
the mean/average.
MeanSquaredError is the default Cost for Annex. MeanSquaredError enables a neural network to more quickly "learn" by associating a higher error penalty for higher error values. Basically, the more incorrect a prediction the higher the penalty by a power of 2.
Link to this section Summary
Functions
Callback implementation for Annex.Cost.calculate/1
.
Link to this section Functions
calculate(error)
View Source
calculate(Annex.Data.List1D.t()) :: float()
calculate(Annex.Data.List1D.t()) :: float()
Callback implementation for Annex.Cost.calculate/1
.