Scheduled Sampling Probability

Problem

Compute the teacher-forcing probability at a training step: clamp(initial_prob + step * rate, 0, 1).

Return the clamped probability. As training progresses the model relies more on its own predictions.

Hint

A negative rate simulates the reverse schedule, decaying toward zero.

case 1

in: [0.1,2,0.1]

expected: 0.30000000000000004

case 2

in: [0.9,5,0.1]

expected: 1

case 3

in: [0,0,0.2]

expected: 0

case 4

in: [0.5,10,0]

expected: 0.5

Starter codePython

def scheduled_sampling_prob(initial_prob, step, rate):
    # Your code here
    pass

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