52 lines
1.1 KiB
Nix
52 lines
1.1 KiB
Nix
{ lib
|
|
, buildPythonPackage
|
|
, fetchFromGitHub
|
|
, linear_operator
|
|
, scikit-learn
|
|
, torch
|
|
, pytestCheckHook
|
|
}:
|
|
|
|
buildPythonPackage rec {
|
|
pname = "gpytorch";
|
|
version = "1.10";
|
|
format = "pyproject";
|
|
|
|
src = fetchFromGitHub {
|
|
owner = "cornellius-gp";
|
|
repo = pname;
|
|
rev = "v${version}";
|
|
hash = "sha256-KY3ItkVjBfIYMkZAmD56EBGR9YN/MRN7b2K3zrK6Qmk=";
|
|
};
|
|
|
|
postPatch = ''
|
|
substituteInPlace setup.py \
|
|
--replace 'find_version("gpytorch", "version.py")' \"$version\"
|
|
'';
|
|
|
|
propagatedBuildInputs = [
|
|
linear_operator
|
|
scikit-learn
|
|
torch
|
|
];
|
|
|
|
checkInputs = [
|
|
pytestCheckHook
|
|
];
|
|
pythonImportsCheck = [ "gpytorch" ];
|
|
disabledTests = [
|
|
# AssertionError on number of warnings emitted
|
|
"test_deprecated_methods"
|
|
# flaky numerical tests
|
|
"test_classification_error"
|
|
"test_matmul_matrix_broadcast"
|
|
];
|
|
|
|
meta = with lib; {
|
|
description = "A highly efficient and modular implementation of Gaussian Processes, with GPU acceleration";
|
|
homepage = "https://gpytorch.ai";
|
|
license = licenses.mit;
|
|
maintainers = with maintainers; [ veprbl ];
|
|
};
|
|
}
|