Exponential LR scheduler

Background

ExponentialLR decays the learning rate by a constant multiplicative factor $\gamma$ at every epoch — a smooth geometric decay, in contrast to StepLR's discrete drops. It is a common, simple schedule when you want the rate to taper continuously over training.

Problem statement

Implement an ExponentialLRScheduler class with:

• __init__(self, initial_lr, gamma)
• get_lr(self, epoch) returning the learning rate at a given (0-indexed) epoch:

$$\text{lr}(e) = \text{lr}_0 \cdot \gamma^{\,e}$$

Round the result to 4 decimals.

Input

• initial_lr — float, the starting learning rate $\text{lr}_0$.
• gamma — float, the per-epoch decay factor (e.g. 0.9 reduces the rate 10% each epoch).
• epoch — int, the current epoch (0-indexed) passed to get_lr.

Output

get_lr(epoch) returns a float learning rate, rounded to 4 decimals.

Examples

Example 1

Input:  ExponentialLRScheduler(initial_lr=0.1, gamma=0.9)
        get_lr(0), get_lr(1), get_lr(2), get_lr(3)
Output: 0.1, 0.09, 0.081, 0.0729

Explanation: each epoch multiplies the previous rate by $\gamma=0.9$: $0.1 \to 0.09 \to 0.081 \to 0.0729$.

Constraints

• The exponent is the epoch itself (decay applies every epoch, not in blocks).
• Round the returned learning rate to 4 decimals.

Notes

• ExponentialLR is StepLR with step_size = 1: the rate drops by $\gamma$ on every epoch.
• Geometric decay means the rate approaches (but never reaches) zero — useful for a long, gentle annealing tail.