Switching to CI to use mamba.

[gomezzz]

Description

Switchin CI to use mamba as tests time out.

Resolved Issues

N/A

How Has This Been Tested?

• CI

Related Pull Requests

N/A

[gomezzz]

@FHof I am switching the CI to mamba from conda but I am getting some errors related to autoray and numpy. Have you maybe encountered them before or an idea why they come up now? 🥲

FAILED boole_test.py::test_integrate_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED gradient_test.py::test_gradients_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED integration_grid_test.py::test_integration_grid_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED integrator_types_test.py::test_integrate_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED monte_carlo_test.py::test_integrate_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED rng_test.py::test_rng_tensorflow_f32 - AttributeError: module 'numpy' has no attribute 'object'
FAILED rng_test.py::test_rng_tensorflow_f64 - AttributeError: module 'numpy' has no attribute 'object'
FAILED simpson_test.py::test_integrate_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED trapezoid_test.py::test_integrate_tensorflow - AttributeError: module 'numpy' has no attribute 'object'
FAILED utils_integration_test.py::test_linspace_with_grads - ImportError: autoray couldn't find function 'array' for backend 'tensorflow'.
FAILED utils_integration_test.py::test_setup_integration_domain - ImportError: autoray couldn't find function 'array' for backend 'tensorflow'.

[FHof]

The tests fail only with TensorFlow (see this part of the build log), and there is a warning tensorflow 2.6.2 requires numpy~=1.19.2, but you have numpy 1.24.1 which is incompatible..
Mamba installs TensorFlow 2.6.2 and the latest release is TensorFlow 2.11.0, and I assume with your latest commit, pip installs a new NumPy version together with JAX, so the new NumPy is loaded instead of the older one installed by mamba.

Without the latest commit (version 06e6c0c), the tests fail only with JAX.
I think the micromamba JAX installation somehow is wrong: It apparently downloads all JAX versions from 0.3.14 down to 0.2.22 and then installs the oldest one, and jaxlib is missing even if I add it explicitly to environment_all_backends.yml.

I've tried some installation configurations. The CI works if TensorFlow is version 2.10.0 and the CPU-only versions of JAX, PyTorch and TensorFlow are installed. With the CUDA versions of TensorFlow or PyTorch, micromamba shows the nothing provides __cuda error. I don't know why this error only appears with a new TensorFlow version.

[gomezzz]

FHof@8f5746f works! Thanks so much @FHof for taking the time 🙏