Trust Region Gain Ratio

Problem

Evaluate the gain ratio of a trust-region step.

If predicted <= 0 use ratio 0.0, otherwise ratio = actual / predicted. Return [ratio, accepted, shrink] where accepted is ratio > eta and shrink is ratio < 0.25.

Hint

A poor ratio means the model predicted more reduction than reality delivered.

Test cases

case 1

in: [0.1,1,0.1]

expected: [0.1,false,true]

case 2

in: [0.9,1,0.1]

expected: [0.9,true,false]

case 3

in: [0.2,0,0.1]

expected: [0,false,true]

case 4

in: [0.05,1,0.25]

expected: [0.05,false,true]

Starter codePython

def trust_region_ratio(actual, predicted, eta):
    # Returns [ratio, accepted, shrink]
    # Your code here
    pass

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