📘

StepLR Scheduler

Problem Statement

Use StepLR to decay the learning rate by a factor every N epochs.

StepLR(optimizer, step_size, gamma) multiplies LR by gamma every step_size epochs. It's the simplest scheduling strategy.

The starter code creates a model and SGD optimizer. Create a StepLR scheduler that halves the learning rate every 3 epochs.

The epoch loop that records LR history is pre-filled.

Returns a dictionary with "lr_history" (10 values), "initial_lr", and "final_lr".

Input:

None

Output:

{'lr_history': [0.1, 0.1, 0.1, 0.05, 0.05, 0.05, 0.025, 0.025, 0.025, 0.0125], 'initial_lr': 0.1, 'final_lr': 0.0125}

Reasoning:

We initialize the learning rate to 0.1 and create a StepLR scheduler with step_size=3 and gamma=0.5, meaning the learning rate will be multiplied by $0.5$ every 3 epochs.
For the first 3 epochs, the learning rate remains at 0.1, as the scheduler hasn't reached its first step.
At epoch 3, the scheduler steps and multiplies the learning rate by $0.5$ , resulting in a new learning rate of $0.1 \cdot 0.5 = 0.05$ , which remains for the next 2 epochs.
This process repeats, with the learning rate being multiplied by $0.5$ every 3 epochs, resulting in the sequence: 0.1, 0.1, 0.1, 0.05, 0.05, 0.05, 0.025, 0.025, 0.025, 0.0125.

Editor

Python 3.13.1

0/0

Run code to see test results.

📘