Passive-Aggressive Update

Problem

Take one passive-aggressive update.

With margin loss = 1 - y * (w dot x), return w unchanged when the loss is non-positive. Otherwise return w + tau * y * x with tau = loss / (||x|| ** 2 + 1 / (2C)).

Hint

The step size tau adapts to the feature norm and the aggressiveness C.

Test cases

case 1

in: [[0,0],[1,1],1,1]

expected: [0.4,0.4]

case 2

in: [[1,1],[1,1],1,0.5]

expected: [1,1]

case 3

in: [[0.2,-0.2],[1,2],-1,2]

expected: [0.047619047619047616,-0.5047619047619047]

Starter codePython

def passive_aggressive_update(w, x, y, C):
    # Your code here
    pass

3 test cases · run your code against them in the editor