2025-07-24 13:55:40 +02:00

88 lines
2.0 KiB
Nix

{
lib,
stdenv,
buildPythonPackage,
fetchFromGitHub,
gpytorch,
linear-operator,
multipledispatch,
pyre-extensions,
pyro-ppl,
setuptools,
setuptools-scm,
torch,
scipy,
pytestCheckHook,
pythonAtLeast,
}:
buildPythonPackage rec {
pname = "botorch";
version = "0.14.0";
pyproject = true;
src = fetchFromGitHub {
owner = "pytorch";
repo = "botorch";
tag = "v${version}";
hash = "sha256-IyRi5kXePnDv2q6SrXLtdltQ1/2/zQ3EBx5phtuX8sE=";
};
build-system = [
setuptools
setuptools-scm
];
dependencies = [
gpytorch
linear-operator
multipledispatch
pyre-extensions
pyro-ppl
scipy
torch
];
nativeCheckInputs = [
pytestCheckHook
];
disabledTests = [
"test_all_cases_covered"
# Skip tests that take too much time
"TestQMultiObjectivePredictiveEntropySearch"
"TestQPredictiveEntropySearch"
]
++ lib.optionals (pythonAtLeast "3.13") [
# RuntimeError: Boolean value of Tensor with more than one value is ambiguous
"test_optimize_acqf_mixed_binary_only"
]
++ lib.optionals (stdenv.buildPlatform.system == "x86_64-linux") [
# stuck tests on hydra
"test_moo_predictive_entropy_search"
]
++ lib.optionals (stdenv.hostPlatform.isDarwin) [
# RuntimeError: required keyword attribute 'value' has the wrong type
"test_posterior_in_trace_mode"
]
++ lib.optionals (stdenv.hostPlatform.isDarwin && stdenv.hostPlatform.isAarch64) [
# Numerical error slightly above threshold
# AssertionError: Tensor-likes are not close!
"test_model_list_gpytorch_model"
];
pythonImportsCheck = [ "botorch" ];
# needs lots of undisturbed CPU time or prone to getting stuck
requiredSystemFeatures = [ "big-parallel" ];
meta = {
changelog = "https://github.com/pytorch/botorch/blob/${src.tag}/CHANGELOG.md";
description = "Bayesian Optimization in PyTorch";
homepage = "https://botorch.org";
license = lib.licenses.mit;
maintainers = with lib.maintainers; [ veprbl ];
};
}