Skip to content

add lazy imports to dataset_utils.py #9052

add lazy imports to dataset_utils.py

add lazy imports to dataset_utils.py #9052

Workflow file for this run

name: Unittests
on:
workflow_dispatch:
pull_request:
branches:
- master
# Do not trigger tests for documentation or markdown docs.
paths-ignore:
- 'docs/**'
- '*.md'
push:
branches:
- master
# Do not trigger tests for documentation or markdown docs.
paths-ignore:
- 'docs/**'
- '*.md'
schedule:
# Trigger tests every day at 02:00 UTC to refresh cache.
- cron: '0 2 * * *'
# Cancel in-progress runs for the current workflow if not on the main branch
# (as it mark the unittests as failed).
# Conditionals to concurrent are based on the solution proposed in this link:
# https://github.community/t/concurrency-cancel-in-progress-but-not-when-ref-is-master/194707
concurrency:
group: ${{ github.workflow }}-${{ github.ref }}-${{ github.ref != 'refs/heads/master' || github.run_number }}
# Cancel only PR intermediate builds
cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
env:
PYTEST_NUM_SHARDS: 4 # Controls tests sharding enabled by `pytest-shard`
jobs:
activate-tests:
name: Check if tests should be run
runs-on: ubuntu-latest
steps:
- name: Check
id: check
# For merged PR, activate testing only on the master branch, based on:
# https://github.community/t/trigger-workflow-only-on-pull-request-merge/17359
run: |
echo "status=${{ github.ref == 'refs/heads/master' || (
github.event.action != 'closed'
&& github.event.pull_request.merged == false
) }}" >> $GITHUB_OUTPUT
outputs:
status: ${{ steps.check.outputs.status }}
shards-job:
needs: activate-tests
if: ${{ needs.activate-tests.outputs.status }}
name: Generate shards
runs-on: ubuntu-latest
steps:
- name: Create variables
id: create-vars
run: |
echo "num-shards=$(jq -n -c '[${{ env.PYTEST_NUM_SHARDS }}]')" >> $GITHUB_OUTPUT
echo "shard-ids=$(jq -n -c '[range(1;${{ env.PYTEST_NUM_SHARDS }}+1)]')" >> $GITHUB_OUTPUT
outputs:
num-shards: ${{ steps.create-vars.outputs.num-shards }}
shard-ids: ${{ steps.create-vars.outputs.shard-ids }}
pytest-job:
needs: shards-job
name: '[${{ matrix.os-version }}][${{ matrix.tf-version }}][Python ${{ matrix.python-version }}][${{ matrix.shard-id }}/${{ matrix.num-shards }}] Core TFDS tests'
runs-on: ${{ matrix.os-version }}
timeout-minutes: 30
strategy:
# Do not cancel in-progress jobs if any matrix job fails.
fail-fast: false
matrix:
tf-version: ['tensorflow']
# Can't reference env variables in matrix
num-shards: ${{ fromJson(needs.shards-job.outputs.num-shards) }}
shard-id: ${{ fromJson(needs.shards-job.outputs.shard-ids) }}
# TF suppported versions: https://www.tensorflow.org/install/pip#software_requirements
python-version: ['3.10', '3.11', '3.12']
os-version: [ubuntu-latest]
steps:
- uses: actions/checkout@v3
- uses: ./.github/actions/setup
with:
tf-version: ${{ matrix.tf-version }}
python-version: ${{ matrix.python-version }}
# Run tests
# Ignores:
# * Nsynth is run in isolation due to dependency conflict (crepe).
# * Lsun tests is disabled because the tensorflow_io used in open-source
# is linked to static libraries compiled again specific TF version, which
# makes test fails with linking error (libtensorflow_io_golang.so).
# * imagenet2012_corrupted requires imagemagick binary.
# * import_without_tf_test.py, because the test relies on TensorFlow not being imported.
# * github_api is run separately to not overuse API quota.
# * wmt is run separately to avoid worker hanging.
# * Huggingface requires `datasets` library.
- name: Run core tests
run: |
pytest --durations=100 -vv -n auto --shard-id=$((${{ matrix.shard-id }} - 1)) --num-shards=${{ env.PYTEST_NUM_SHARDS }} \
--ignore="tensorflow_datasets/datasets/nsynth/nsynth_dataset_builder_test.py" \
--ignore="tensorflow_datasets/image/lsun_test.py" \
--ignore="tensorflow_datasets/datasets/imagenet2012_corrupted/imagenet2012_corrupted_dataset_builder_test.py" \
--ignore="tensorflow_datasets/scripts/documentation/build_api_docs_test.py" \
--ignore="tensorflow_datasets/import_without_tf_test.py" \
--ignore="tensorflow_datasets/core/github_api/github_path_test.py" \
--ignore="tensorflow_datasets/translate/wmt19_test.py" \
--ignore="tensorflow_datasets/core/dataset_builders/huggingface_dataset_builder_test.py" \
--ignore="tensorflow_datasets/core/utils/huggingface_utils_test.py"
# Run tests without any pytest plugins. The tests should be triggered for a single shard only.
- name: Run leftover tests
if: ${{ matrix.shard-id == 1 }}
uses: nick-fields/retry@v2
with:
timeout_minutes: 1
max_attempts: 2
retry_on: timeout
command: |
pytest -vv -o faulthandler_timeout=10 tensorflow_datasets/translate/wmt19_test.py
huggingface-pytest-job:
needs: activate-tests
if: ${{ needs.activate-tests.outputs.status }}
# HuggingFace tests need to be run separately because they're disabled without installed
# `datasets` library.
name: 'HuggingFace Python 3.10 tests'
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- uses: actions/checkout@v3
- uses: ./.github/actions/setup
with:
tf-version: tensorflow
python-version: '3.10'
extras: huggingface
- name: Run HuggingFace tests
run: |
pytest -vv -n auto \
tensorflow_datasets/core/dataset_builders/huggingface_dataset_builder_test.py \
tensorflow_datasets/core/utils/huggingface_utils_test.py
githubapi-pytest-job:
needs: activate-tests
if: ${{ needs.activate-tests.outputs.status }}
name: 'Github API tests'
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- uses: actions/checkout@v3
- uses: ./.github/actions/setup
with:
tf-version: tensorflow
- name: Run Github API tests
run: pytest --durations=100 -vv -n auto tensorflow_datasets/core/github_api/github_path_test.py
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
notebook-test-job:
needs: activate-tests
if: ${{ needs.activate-tests.outputs.status }}
name: 'Notebook tests'
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- uses: actions/checkout@v3
- uses: ./.github/actions/setup
with:
tf-version: tensorflow
use-cache: false
# Test each notebook sequentially.
- name: Run notebook
run: |
ipython kernel install --user --name tfds-notebook
for notebook in docs/*ipynb
do
# These notebooks time out because they rely on loading huge datasets.
if [[ "$notebook" != "docs/determinism.ipynb" ]] && \
[[ "$notebook" != "docs/dataset_collections.ipynb" ]]
then
jupyter nbconvert \
--ExecutePreprocessor.timeout=600 \
--ExecutePreprocessor.kernel_name=tfds-notebook \
--to notebook \
--execute $notebook && \
pip install tensorflow # reinstall tensorflow if it was uninstalled
fi
done