Lookahead Multi-Step Averaging

Problem

Apply several Lookahead interpolations to a slow weight vector.

Starting from slow, process each fast vector in fast_history in order with slow = slow + alpha * (fast - slow). Return [slow_final, fast_final] where fast_final is a copy of the final slow weights. An empty history returns the starting weights unchanged.

Hint

Each interpolation moves the slow weights a fraction alpha toward the observed fast weights.

Test cases

case 1

in: [[0,0],[[2,4],[3,1]],0.5]

expected: [[2,1.5],[2,1.5]]

case 2

in: [[1,1],[[1,1]],0.9]

expected: [[1,1],[1,1]]

case 3

in: [[0],[[10],[0],[10]],0.25]

expected: [[3.90625],[3.90625]]

Starter codePython

def lookahead_multistep(slow, fast_history, alpha):
    # Returns [slow_final, fast_final]
    # Your code here
    pass

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