dtorch.loss

Description

A loss function is generaly a way to establish how much errors a model is doing. Commons loss function are present in this package.

It’s usage is pretty straightforward. Ex: .. code::python

>> loss = dtorch.loss.MSELoss() >> input = dtorch.random(3, 5, requires_grad=True) >> target = dtorch.random(3, 5) >> output = loss(input, target) >> output.backward()

Loss

class MSELoss

Mean Squared Error is a function that is the mean of the squared residual between two set of data.

Mathematicaly: \(y = \frac{1}{n} \sum_{i=1}^N(y_i - ŷ_i)^2\) where ŷ is the wanted result.

class BCELoss

Warning

It is not tested yet

BCELoss is a way to evaluate how good a prediction is between two classes. It’s used for binary classification as such.

Mathematicaly it can be written as: \(-\frac{1}{n}\sum_{i=0}^Ny_i*log(ŷ_i)+(1-y_i)*log(1-ŷ_i)\)