Benchmarks

Benchmarks in CL-Gym.

Benchmark(num_tasks, per_task_examples, …)

Base class for continual learning benchmarks.

mnist.ContinualMNIST(num_tasks, …)

Base class for (Permuted/Rotated/Split)-MNIST benchmarks.

RotatedMNIST(num_tasks, per_task_examples, …)

Rotated MNIST benchmark.

PermutedMNIST(num_tasks, per_task_examples, …)

Permuted MNIST benchmark

SplitCIFAR10(num_tasks, per_task_examples, …)

Split CIFAR-10 benchmark.

SplitCIFAR100(num_tasks, per_task_examples, …)

Split CIFAR-100 benchmark.

Toy2DClassification(num_tasks, …)

Toy benchmark: each task will be a binary classification with linearly separable classes in 2D space.

Toy1DRegression(num_tasks, …)

Toy benchmark: 1D Regression problem with 3 tasks Task t will be a polynomial with degree t Please see: https://openreview.net/pdf?id=SJgwNerKvB

PAMAP2(num_tasks, per_task_examples, …[, …])

PAMAP2 benchmark: activity recognition with at most 8 tasks.
How does CL-Gym work? Benchmark