Machine learning framework PaddlePaddle loss function and optimization method of the component

First, the loss function

1, the mean absolute error loss function (L1)

Definition: the absolute value of the average of the predicted value of the target before the

2, the mean square error loss function (L2)

Definition: the mean of the squared difference between the predicted value and the target value

 

loss = paddle.fluid.layers.square_error_cost(input, label)

 

paddle/fluid/layers/nn.py

 

 

3, the smoothing function the mean absolute error (Huber)

Definition: the predicted value and the target value when the difference is large, an average absolute error; when the difference between the predicted value and the target value is small, the use of mean square error

4, the hyperbolic cosine function loss (Log-Cosh)

5, quantile loss function (Quantile)

Definition: Different embodiments penalties for positive and negative difference between the predicted value and the target value

Second, the optimization method