depot/third_party/nixpkgs/pkgs/development/python-modules/tensorflow-probability/default.nix

135 lines
3 KiB
Nix

{
lib,
stdenv,
# bazel wheel
buildBazelPackage,
fetchFromGitHub,
# nativeBuildInputs
python,
setuptools,
wheel,
absl-py,
bazel_6,
cctools,
# python package
buildPythonPackage,
# dependencies
cloudpickle,
decorator,
dm-tree,
gast,
keras,
numpy,
six,
tensorflow,
# tests
hypothesis,
matplotlib,
mock,
mpmath,
pandas,
pytest,
scipy,
}:
let
version = "0.25.0";
pname = "tensorflow-probability";
# first build all binaries and generate setup.py using bazel
bazel-wheel = buildBazelPackage {
name = "tensorflow_probability-${version}-py2.py3-none-any.whl";
src = fetchFromGitHub {
owner = "tensorflow";
repo = "probability";
rev = "refs/tags/v${version}";
hash = "sha256-LXQfGFgnM7WYUQjJ2Y3jskdeJ/dEKz+Afg+UOQjv5kc=";
};
nativeBuildInputs = [
absl-py
# needed to create the output wheel in installPhase
python
setuptools
tensorflow
wheel
];
bazel = bazel_6;
bazelTargets = [ ":pip_pkg" ];
LIBTOOL = lib.optionalString stdenv.hostPlatform.isDarwin "${cctools}/bin/libtool";
fetchAttrs = {
sha256 = "sha256-TbWcWYidyXuAMgBnO2/k0NKCzc4wThf2uUeC3QxdBJY=";
};
buildAttrs = {
preBuild = ''
patchShebangs .
'';
installPhase = ''
# work around timestamp issues
# https://github.com/NixOS/nixpkgs/issues/270#issuecomment-467583872
export SOURCE_DATE_EPOCH=315532800
# First build, then move. Otherwise pip_pkg would create the dir $out
# and then put the wheel in that directory. However we want $out to
# point directly to the wheel file.
./bazel-bin/pip_pkg . --release
mv *.whl "$out"
'';
};
};
in
buildPythonPackage {
inherit version pname;
format = "wheel";
src = bazel-wheel;
dependencies = [
cloudpickle
decorator
dm-tree
gast
keras
numpy
six
tensorflow
];
# Listed here:
# https://github.com/tensorflow/probability/blob/f3777158691787d3658b5e80883fe1a933d48989/testing/dependency_install_lib.sh#L83
nativeCheckInputs = [
hypothesis
matplotlib
mock
mpmath
pandas
pytest
scipy
];
# Ideally, we run unit tests with pytest, but in checkPhase, only the Bazel-build wheel is available.
# But it seems not guaranteed that running the tests with pytest will even work, see
# https://github.com/tensorflow/probability/blob/c2a10877feb2c4c06a4dc58281e69c37a11315b9/CONTRIBUTING.md?plain=1#L69
# Ideally, tests would be run using Bazel. For now, lets's do a...
# sanity check
pythonImportsCheck = [ "tensorflow_probability" ];
meta = {
description = "Library for probabilistic reasoning and statistical analysis";
homepage = "https://www.tensorflow.org/probability/";
changelog = "https://github.com/tensorflow/probability/releases/tag/v${version}";
license = lib.licenses.asl20;
maintainers = with lib.maintainers; [ GaetanLepage ];
};
}