python3Packages.torcheval: skip failing tests (#455334)

{

  lib,

  stdenv,

  buildPythonPackage,

  fetchFromGitHub,

  # dependencies

  typing-extensions,

  # tests

  cython,

  numpy,

  pytest-timeout,

  pytest-xdist,

  pytestCheckHook,

  scikit-image,

  scikit-learn,

  torchtnt-nightly,

  torchvision,

}:

let

  pname = "torcheval";

  pyproject = true;

  src = fetchFromGitHub {

    owner = "pytorch";

    owner = "meta-pytorch";

    repo = "torcheval";

    # Upstream has not created a tag for this version

    # https://github.com/pytorch/torcheval/issues/215

    hash = "sha256-aVr4qKKE+dpBcJEi1qZJBljFLUl8d7D306Dy8uOojJE=";

  };

  # Patches are only applied to usages of numpy within tests,

  # which are only used for testing purposes (see dev-requirements.txt)

  postPatch =

    # numpy's `np.NAN` was changed to `np.nan` when numpy 2 was released

    ''

      substituteInPlace tests/metrics/classification/test_accuracy.py tests/metrics/functional/classification/test_accuracy.py \

        --replace-fail "np.NAN" "np.nan"

    ''

    # `unittest.TestCase.assertEquals` does not exist;

    # the correct symbol is `unittest.TestCase.assertEqual`

    + ''

      substituteInPlace tests/metrics/test_synclib.py \

        --replace-fail "tc.assertEquals" "tc.assertEqual"

    '';

  build-system = [ setuptools ];

  dependencies = [ typing-extensions ];

  pythonImportsCheck = [ "torcheval" ];

  nativeCheckInputs = [

    cython

    numpy

    pytest-timeout

    pytest-xdist

    pytestCheckHook

    scikit-image

    scikit-learn

    torchtnt-nightly

    torchvision

  ];

  pytestFlags = [

    "-v"

  ];

  enabledTestPaths = [

    "tests/"

  ];

  disabledTestPaths = [

    # -- tests/metrics/audio/test_fad.py --

    # Touch filesystem and require network access.

    # torchaudio.utils.download_asset("models/vggish.pt") -> PermissionError: [Errno 13] Permission denied: '/homeless-shelter'

    "tests/metrics/audio/test_fad.py::TestFAD::test_vggish_fad"

    "tests/metrics/audio/test_fad.py::TestFAD::test_vggish_fad_merge"

    # -- tests/metrics/image/test_fid.py --

    # Touch filesystem and require network access.

    # models.inception_v3(weights=weights) -> PermissionError: [Errno 13] Permission denied: '/homeless-shelter'

    "tests/metrics/image/test_fid.py::TestFrechetInceptionDistance::test_fid_invalid_input"

    "tests/metrics/image/test_fid.py::TestFrechetInceptionDistance::test_fid_random_data_custom_model"

    "tests/metrics/image/test_fid.py::TestFrechetInceptionDistance::test_fid_random_data_default_model"

    "tests/metrics/image/test_fid.py::TestFrechetInceptionDistance::test_fid_with_dissimilar_inputs"

    "tests/metrics/image/test_fid.py::TestFrechetInceptionDistance::test_fid_with_similar_inputs"

    # -- tests/metrics/functional/text/test_perplexity.py --

    # AssertionError: Scalars are not close!

    # Expected 3.537154912949 but got 3.53715443611145

    "tests/metrics/functional/text/test_perplexity.py::Perplexity::test_perplexity_with_ignore_index"

    # -- tests/metrics/image/test_psnr.py --

    # AssertionError: Scalars are not close!

    # Expected 7.781850814819336 but got 7.781772613525391

    "tests/metrics/image/test_psnr.py::TestPeakSignalNoiseRatio::test_psnr_with_random_data"

    # -- tests/metrics/regression/test_mean_squared_error.py --

    # AssertionError: Scalars are not close!

    # Expected -640.4547729492188 but got -640.4707641601562

    "tests/metrics/regression/test_mean_squared_error.py::TestMeanSquaredError::test_mean_squared_error_class_update_input_shape_different"

    # -- tests/metrics/window/test_mean_squared_error.py --

    # AssertionError: Scalars are not close!

    # Expected 0.0009198983898386359 but got 0.0009198188781738281

    "tests/metrics/window/test_mean_squared_error.py::TestMeanSquaredError::test_mean_squared_error_class_update_input_shape_different"

  ]

  # These tests error on darwin platforms.

  # NotImplementedError: The operator 'c10d::allgather_' is not currently implemented for the mps device

  #

  # Applying the suggested environment variable `PYTORCH_ENABLE_MPS_FALLBACK=1;` causes the tests to fail,

  # as using the CPU instead of the MPS causes the tensors to be on the wrong device:

  # RuntimeError: ProcessGroupGloo::allgather: invalid tensor type at index 0;

  # Expected TensorOptions(dtype=float, device=cpu, ...), got TensorOptions(dtype=float, device=mps:0, ...)

  ++ lib.optional stdenv.hostPlatform.isDarwin [

    # -- tests/metrics/test_synclib.py --

    "tests/metrics/test_synclib.py::SynclibTest::test_complex_mixed_state_sync"

    "tests/metrics/test_synclib.py::SynclibTest::test_complex_mixed_state_sync"

    "tests/metrics/test_synclib.py::SynclibTest::test_empty_tensor_list_sync_state"

    "tests/metrics/test_synclib.py::SynclibTest::test_sync_dtype_and_shape"

    "tests/metrics/test_synclib.py::SynclibTest::test_tensor_list_sync_states"

    "tests/metrics/test_synclib.py::SynclibTest::test_tensor_dict_sync_states"

    "tests/metrics/test_synclib.py::SynclibTest::test_tensor_sync_states"

    # -- tests/metrics/test_toolkit.py --

    "tests/metrics/test_toolkit.py::MetricToolkitTest::test_metric_sync"

    "tests/metrics/test_toolkit.py::MetricCollectionToolkitTest::test_metric_collection_sync"

    # Cannot access local process over IPv6 (nodename nor servname provided) even with __darwinAllowLocalNetworking

    # Will hang, or appear to hang, with an 5 minute (default) timeout per test

    "tests/metrics/aggregation/test_auc.py"

    "tests/metrics/aggregation/test_cat.py"

    "tests/metrics/aggregation/test_max.py"

    "tests/metrics/aggregation/test_mean.py"

    "tests/metrics/aggregation/test_min.py"

    "tests/metrics/aggregation/test_sum.py"

    "tests/metrics/aggregation/test_throughput.py"

    "tests/metrics/classification/test_accuracy.py"

    "tests/metrics/classification/test_auprc.py"

    "tests/metrics/classification/test_auroc.py"

    "tests/metrics/classification/test_binned_auprc.py"

    "tests/metrics/classification/test_binned_auroc.py"

    "tests/metrics/classification/test_binned_precision_recall_curve.py"

    "tests/metrics/classification/test_confusion_matrix.py"

    "tests/metrics/classification/test_f1_score.py"

    "tests/metrics/classification/test_normalized_entropy.py"

    "tests/metrics/classification/test_precision_recall_curve.py"

    "tests/metrics/classification/test_precision.py"

    "tests/metrics/classification/test_recall_at_fixed_precision.py"

    "tests/metrics/classification/test_recall.py"

    "tests/metrics/functional/classification/test_auroc.py"

    "tests/metrics/ranking/test_click_through_rate.py::TestClickThroughRate::test_ctr_with_valid_input"

    "tests/metrics/ranking/test_hit_rate.py::TestHitRate::test_hitrate_with_valid_input"

    "tests/metrics/ranking/test_reciprocal_rank.py::TestReciprocalRank::test_mrr_with_valid_input"

    "tests/metrics/ranking/test_retrieval_precision.py::TestRetrievalPrecision::test_retrieval_precision_multiple_updates_1_query"

    "tests/metrics/ranking/test_retrieval_precision.py::TestRetrievalPrecision::test_retrieval_precision_multiple_updates_n_queries_without_nan"

    "tests/metrics/ranking/test_weighted_calibration.py::TestWeightedCalibration::test_weighted_calibration_with_valid_input"

    "tests/metrics/regression/test_mean_squared_error.py"

    "tests/metrics/regression/test_r2_score.py"

    "tests/metrics/test_synclib.py::SynclibTest::test_gather_uneven_multidim"

    "tests/metrics/test_synclib.py::SynclibTest::test_gather_uneven"

    "tests/metrics/test_synclib.py::SynclibTest::test_numeric_sync_state"

    "tests/metrics/test_synclib.py::SynclibTest::test_sync_list_length"

    "tests/metrics/text/test_bleu.py::TestBleu::test_bleu_multiple_examples_per_update"

    "tests/metrics/text/test_bleu.py::TestBleu::test_bleu_multiple_updates"

    "tests/metrics/text/test_perplexity.py::TestPerplexity::test_perplexity_with_ignore_index"

    "tests/metrics/text/test_perplexity.py::TestPerplexity::test_perplexity"

    "tests/metrics/text/test_word_error_rate.py::TestWordErrorRate::test_word_error_rate_with_valid_input"

    "tests/metrics/text/test_word_information_lost.py::TestWordInformationLost::test_word_information_lost"

    "tests/metrics/text/test_word_information_preserved.py::TestWordInformationPreserved::test_word_information_preserved_with_valid_input"

    "tests/metrics/window/test_auroc.py"

    "tests/metrics/window/test_click_through_rate.py::TestClickThroughRate::test_ctr_with_valid_input"

    "tests/metrics/window/test_mean_squared_error.py"

    "tests/metrics/window/test_normalized_entropy.py::TestWindowedBinaryNormalizedEntropy::test_ne_with_valid_input"

    "tests/metrics/window/test_weighted_calibration.py::TestWindowedWeightedCalibration::test_weighted_calibration_with_valid_input"

  ];

  # Tests are very flaky and computationally intensive

  doCheck = false;

  meta = {

    description = "Rich collection of performant PyTorch model metrics and tools for PyTorch model evaluations";

    homepage = "https://pytorch.org/torcheval";

    changelog = "https://github.com/pytorch/torcheval/releases/tag/${version}";

    changelog = "https://github.com/meta-pytorch/torcheval/releases/tag/${version}";

    platforms = lib.platforms.unix;

    license = [ lib.licenses.bsd3 ];