Adding custom loss functions

Back to Home

Loss functions are stored in an Object value named lossfuncs.
For nodejs require the object like so:

const lossfuncs = require('dannjs').lossfuncs;

To add your own function, add a new object property with the function you want to add. The name of the property is the name the Dann model is going to use to refer to the function.

lossfuncs.name = function;

Here's an example with the pre-existing 'mse' loss function:

const dn = require('dannjs');     //nodejs only
const lossfuncs = dn.lossfuncs;   //nodejs only

lossfuncs.mse = function mse(predictions,target) {
    let sum = 0;
    let n = target.length;

    for (let i = 0; i < n; i++) {
        let y = target[i];
        let yHat = predictions[i];
        sum += Math.pow(y - yHat,2);
    }

    return sum/n;
}

Function arguments

A standard loss function takes two arguments to evaluate the resulting loss of the model.

• predictions
  The predicted output of the model as an array.

• target
  The desired output of the model as an array.

• returns
  A loss value.