PR3: Add cifar10 experiments #171
Conversation
There was a problem hiding this comment.
Mostly cosmetic changes suggested throughout. The biggest set of possible changes are proposed in this PR
tests/utils/sampler_test.py
Outdated
| sampler = DirichletLabelBasedSampler(unique_labels=list(range(10)), sample_percentage=1.0, beta=0.1) | ||
| ds = MnistDataset(data_path=Path("examples/datasets/MNIST"), train=True) |
There was a problem hiding this comment.
We're forming the MnistDataset in a few places and this test will require downloading the dataset if it doesn't exist in the folder specified. So I'm thinking we can just use a synthetic dataset for these tests. I encapsulated the possible simplification in this PR
There was a problem hiding this comment.
Even though this is a comment on outdated code, we're still using the MNIST dataset here. So I think the migration to the synthetic dataset will still be useful unless you disagree 🙂
sanaAyrml
changed the base branch from
sa_add_deep_mmd_loss
to
sa_update_mkmmd_loss
sanaAyrml
changed the base branch from
sa_update_mkmmd_loss
to
sa_add_deep_mmd_loss
sanaAyrml
changed the base branch from
sa_add_deep_mmd_loss
to
sa_update_mkmmd_loss
|
Resolved all feedback and implemented four checkpointing methods for clients and two for the server. Additionally, expanded the test file to evaluate all checkpoint scenarios. |
for more information, see https://pre-commit.ci
…rInstitute/FL4Health into sa_add_cifar10_experiments
|
To other final comments:
|
|
I have made all the necessary changes and added a README to the |
PR Type
[Feature ]
Short Description
Cifar10 experiments with various levels of heterogeneity.
Tests Added
No tests added yet.