Skip to content

Window shuffle set_state leads to wrong window_index. #1178

@markblee

Description

@markblee

Hi, it seems like WindowShuffleDatasetIterator should set self._init = False in set_state. Not doing so seems to produce the wrong window_index when restoring.

Repro:

import grain.python as grain
from grain._src.python.dataset.transformations import shuffle as grain_shuffle

ds = grain.MapDataset.source(grain.RangeDataSource(start=0, stop=10, step=1))
ds = ds.to_iter_dataset()
ds = grain_shuffle.WindowShuffleIterDataset(ds, window_size=5, seed=42)

it1 = iter(ds)
checkpoint = it1.get_state()
original = list(it1)

it2 = iter(ds)
it2.set_state(checkpoint)
restored = list(it2)

assert original == restored, (original, restored)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions