osx_arm updates

.github/workflows/workflow.yml

@@ -222,28 +222,18 @@ jobs:
           retention-days: 7
 
   osx_arm:
-    defaults:
-      run:
-        shell: bash -l {0}
     needs: check
     runs-on: macos-14
     # Run on every code push, but only on review ready PRs
     if: ${{ github.event_name == 'push' || github.event.pull_request.draft == false }}
     steps:
     - uses: actions/checkout@v4
-    - name: Set up Conda
-      uses: conda-incubator/setup-miniconda@v2.3.0
+    - uses: actions/setup-python@v5
       with:
-        miniconda-version: "latest"
-        auto-update-conda: true
-        python-version: '3.8'
-        activate-environment: aspire
-        environment-file: environment-accelerate.yml
-        auto-activate-base: false
-    - name: Complete Install and Log Environment ${{ matrix.os }} Python ${{ matrix.python-version }}
+        python-version: '3.11'
+    - name: Complete Install and Log Environment
       run: |
-        conda info
-        conda list
+        python --version
         pip install -e ".[dev]"  # install aspire
         pip freeze
     - name: Test

README.md

@@ -37,11 +37,7 @@ install `aspire` safely in that environment.
 
 If you are unfamiliar with `conda`, the
 [Miniconda](https://docs.conda.io/en/latest/miniconda.html)
-distribution for `x86_64` is recommended.  For Apple silicon to use
-the osx-arm platform, patching and building some dependencies from
-source is currently required.  The Intel `osx-64` install is still
-preferred even for Apple silicon users, otherwise [notes are
-provided.](https://github.com/ComputationalCryoEM/ASPIRE-Python/discussions/969)
+distribution for `x86_64` is recommended.
 
 Assuming you have `conda` and a compatible system, the following steps
 will checkout current code release, create an environment, and install

docs/source/installation.rst

@@ -34,12 +34,6 @@ to view Conda's installation instructions.
 .. note:: If you're not sure which distribution is right for you, go
    with `Miniconda <https://conda.io/miniconda.html>`__
 
-.. note:: For Apple silicon to use the osx-arm platform, patching and
-   building some dependencies from source is currently required.  The
-   Intel ``osx-64`` install is still preferred even for Apple silicon
-   users, otherwise `notes are
-   provided. <https://github.com/ComputationalCryoEM/ASPIRE-Python/discussions/969>`_
-
 Getting Started - Installation
 ************************************
 

tests/test_FLEbasis2D.py

@@ -1,4 +1,5 @@
 import os
+import platform
 import sys
 
 import numpy as np
@@ -70,8 +71,12 @@ def relerr(base, approx):
 
 @pytest.mark.parametrize("basis", test_bases, ids=show_fle_params)
 class TestFLEBasis2D(UniversalBasisMixin):
-    # Loosen the tolerance for `cufinufft` to be within 15%
-    test_eps = 1.15 if backend_available("cufinufft") else 1.0
+    # Loosen the tolerance for `cufinufft` and `osx_arm`
+    test_eps = 1.0
+    if backend_available("cufinufft"):
+        test_eps = 1.15
+    elif platform.system() == "Darwin":
+        test_eps = 1.20
 
     # check closeness guarantees for fast vs dense matrix method
     def testFastVDense_T(self, basis):

tests/test_indexed_source.py

@@ -22,11 +22,15 @@ def test_remapping(sim_fixture):
     sim, sim2 = sim_fixture
 
     # Check images are served correctly, using internal index.
-    assert np.allclose(sim.images[sim2.index_map].asnumpy(), sim2.images[:].asnumpy())
+    np.testing.assert_allclose(
+        sim.images[sim2.index_map].asnumpy(), sim2.images[:].asnumpy(), atol=1e-6
+    )
 
     # Check images are served correctly, using known index (evens).
     index = list(range(0, sim.n, 2))
-    assert np.allclose(sim.images[index].asnumpy(), sim2.images[:].asnumpy())
+    np.testing.assert_allclose(
+        sim.images[index].asnumpy(), sim2.images[:].asnumpy(), atol=1e-6
+    )
 
     # Check meta is served correctly.
     assert np.all(sim.get_metadata(indices=sim2.index_map) == sim2.get_metadata())

tests/test_preprocess_pipeline.py

@@ -168,7 +168,7 @@ def testInvertContrast(L, dtype):
 
     # all images should be the same after inverting contrast
     np.testing.assert_allclose(
-        imgs1_rc.asnumpy(), imgs2_rc.asnumpy(), rtol=1e-05, atol=1e-08
+        imgs1_rc.asnumpy(), imgs2_rc.asnumpy(), rtol=1e-05, atol=1e-06
     )
     # dtype of returned images should be the same
     assert dtype == imgs1_rc.dtype