%global pypi_name robustica

Name:           python-%{pypi_name}
Version:        0.1.4
Release:        1%{?dist}
Summary:        Robust Independent Component Analysis by clustering repeated FastICA runs

# README and LICENSE say BSD 3-Clause. The LICENSE text is a copy of
# scikit-learn's, with its copyright line; the source files carry only a
# "Centre de Regulacio Genomica - All Rights Reserved" header. Flagged upstream
# in docs/upstream/robustica-licence.md.
License:        BSD-3-Clause
URL:            https://github.com/CRG-CNAG/robustica
Source0:        %{pypi_source %{pypi_name}}

BuildArch:      noarch
BuildRequires:  python3-devel

%global _description %{expand:
RobustICA runs FastICA many times, clusters the resulting components and
returns the cluster centroids with a quality index for each, so that the
components are reproducible rather than an accident of one random start.
tedana uses it as an alternative to a single FastICA run.}

%description %_description

%package -n python3-%{pypi_name}
Summary:        %{summary}

%description -n python3-%{pypi_name} %_description

%prep
%autosetup -n %{pypi_name}-%{version}

%generate_buildrequires
%pyproject_buildrequires

%build
%pyproject_wheel

%install
%pyproject_install
%pyproject_save_files -l robustica

%check
# Four known independent sources (square wave, Laplace noise, a cubed sine,
# uniform noise) mixed into 30 channels with a little noise: RobustICA must
# return all four, each correlated > 0.99 with its source.
cd %{_tmppath}
%{py3_test_envvars} %{python3} - <<'PY'
import numpy as np
from robustica import RobustICA, __version__
assert __version__ == "%{version}"
rng = np.random.default_rng(0)
n = 2000
S = np.stack([np.sign(np.sin(np.linspace(0, 60, n))), rng.laplace(size=n),
              np.sin(np.linspace(0, 31, n))**3, rng.uniform(-1, 1, n)])
X = S.T @ rng.standard_normal((30, 4)).T + 0.05*rng.standard_normal((n, 30))
est, mixing = RobustICA(n_components=4, robust_runs=10, random_state=0,
                        n_jobs=1).fit_transform(X)
assert est.shape == (n, 4) and mixing.shape == (30, 4)
C = np.abs(np.corrcoef(est.T, S)[:4, 4:])
print(C.round(3))
assert (C.max(axis=1) > 0.99).all()
assert sorted(C.argmax(axis=1)) == [0, 1, 2, 3]
PY

%files -n python3-%{pypi_name} -f %{pyproject_files}
%license LICENSE
%doc README.md

%changelog
* Fri Oct 16 2026 Morgan Hough <morgan.hough@gmail.com> - 0.1.4-1
- New package: required by tedana 26