Easy

Learning Rate Decay

Easy

~10 min

code completion

Learning Rate Decay

A fixed learning rate is a compromise: large enough to make fast progress early, small enough to converge precisely later. Exponential decay solves this by shrinking the learning rate over time:

α_{t} = α_{0} \cdot γ^{t}

where:

α_{0}

is the initial learning rate

γ

(gamma/decay_rate) is the decay factor,

0 < γ < 1

t

is the current epoch number

At epoch 0, $α_{t} = α_{0}$ . Each epoch multiplies by $γ$ .

Your task:

Implement decay_lr(initial_lr, decay_rate, epoch) that returns the decayed learning rate.

Example Tests

Epoch 0: no decay

Input: {"epoch":0,"decay_rate":0.9,"initial_lr":0.1}

Expected: 0.1

Epoch 1: single decay step

Input: {"epoch":1,"decay_rate":0.9,"initial_lr":0.1}

Expected: 0.09

Halving decay after 4 epochs

Input: {"epoch":4,"decay_rate":0.5,"initial_lr":1}

Expected: 0.0625

You can read the full problem statement above. Create a free account to run code in the browser, submit solutions, and track your progress.