Macromol-voxelize

Fix

* fix: don&39;t require numpy.typing ([`a4ae67b`](https://github.com/kalekundert/macromol_voxelize/commit/a4ae67bd35a8571db9d5b7d74267810f432557af))

Feature

* feat: support python 3.8

The rosetta docker image still uses python 3.8, so I need to support it. ([`3b2deab`](https://github.com/kalekundert/macromol_voxelize/commit/3b2deab8d0524f4cc54da406bccc9279cc069ef9))

* feat: add outline argument to load_voxels ([`798e687`](https://github.com/kalekundert/macromol_voxelize/commit/798e6875f0c4b8ed725b1d123d6ed3472d983abe))

* feat: view batched images in pymol ([`5ad859f`](https://github.com/kalekundert/macromol_voxelize/commit/5ad859f28d69202979715270b0ea037c2d83c105))

Fix

* fix: use up-to-date image_from_atoms() API in pymol ([`aadc50e`](https://github.com/kalekundert/macromol_voxelize/commit/aadc50e1bdaa19780339d3794ff9c2cd884b041e))

Test

* test: run PyMOL tests on headless CI servers ([`06d217a`](https://github.com/kalekundert/macromol_voxelize/commit/06d217a6198d5d3b44ad374cbc4d9488eafd5889))

Chore

* chore: pin setuptools for PyPy

This is a work-around for pypa/distutils283 ([`ae2bb42`](https://github.com/kalekundert/macromol_voxelize/commit/ae2bb4259c6374782aa6fa1eff388eaa30fceb03))

* chore: remove C++ coverage for CI

I don&39;t think this ever really worked, and now it&39;s been causing the
build to fail for a while. If I ever want to come back to this, I can
come back to the previous commit to get a good starting point. ([`a1cdb8d`](https://github.com/kalekundert/macromol_voxelize/commit/a1cdb8df5f9e4ffade480da107ea6344c7e0b0d2))

* chore: fix upload of coverage artifacts ([`1411f3b`](https://github.com/kalekundert/macromol_voxelize/commit/1411f3beaba7c8c8087b177b649368783e155f84))

Documentation

* docs: update description of how channels are determined ([`8c06062`](https://github.com/kalekundert/macromol_voxelize/commit/8c06062b723f5c3f0c7524fea728c7ffaede2c33))

Feature

* feat: use file name to name pymol voxel objects ([`dc9fd24`](https://github.com/kalekundert/macromol_voxelize/commit/dc9fd24fcf31a638b66ccfda6dc4caea289ae4a7))

* feat: make an API for pre-filtering atoms

This API replaces the `process_filtered_atoms` callback. After living
with this callback for a while, I ultimately decided that it was awkward
to use, and that exposing separate functions for the filtering and
image-making tasks would be more powerful. ([`5d4683b`](https://github.com/kalekundert/macromol_voxelize/commit/5d4683b8b7098c38dd9413770d539268aeebff2c))

* feat: add option to scale alpha when loading voxels ([`833993e`](https://github.com/kalekundert/macromol_voxelize/commit/833993e79c64792490861339078c408218a1dda6))

* feat: use a better default color scheme ([`fb01d3c`](https://github.com/kalekundert/macromol_voxelize/commit/fb01d3c840fb45c7fe1c27f61bb74df76f5bff95))

* feat: add option to save image being visualized ([`eb95c16`](https://github.com/kalekundert/macromol_voxelize/commit/eb95c16229773a597a21f8941b640ef4939942f6))

Fix

* fix: weight atoms by occupancy when picking channel colors ([`6aaade2`](https://github.com/kalekundert/macromol_voxelize/commit/6aaade27331de142a2a6fbc499957fcc63bda81a))

Test

* test: disable flaky test ([`72700b6`](https://github.com/kalekundert/macromol_voxelize/commit/72700b6a170267936ceb6bb60b72e7e52d804660))

Feature

* feat: render images exported directly from numpy ([`441baaa`](https://github.com/kalekundert/macromol_voxelize/commit/441baaa28e8cc71aa03d0b315631238392cf3692))

* feat: make atom channels from polars expressions ([`3bf2829`](https://github.com/kalekundert/macromol_voxelize/commit/3bf282939d880e90d2f00930ef67e3381c5c8217))

* feat: add option to scale the alpha channel ([`9cc9de8`](https://github.com/kalekundert/macromol_voxelize/commit/9cc9de8574cdfc05d8f9e035f1292a7f578ad261))

Fix

* fix: upgrade to polars 1.0 ([`003a78d`](https://github.com/kalekundert/macromol_voxelize/commit/003a78dd2697ff3adf4958f651eb2bcbdc9f1014))

* fix: allow mmvox and ap to be used together ([`cdeb947`](https://github.com/kalekundert/macromol_voxelize/commit/cdeb947e5af630edfde5ac2e4f9888f516e37a73))

* fix: don&39;t use removed atom channels argument ([`ffa837c`](https://github.com/kalekundert/macromol_voxelize/commit/ffa837c89d9b58d699459cd1a3cbf035c50f3fc0))

Feature

* feat: pass atoms dataframe to C++ via numpy

The previous commit introduced code to pass atoms to C++ via Arrow
tables. This would&39;ve been a better solution, because it would never
require copying data. Unfortunately, it requires linking to the arrow
and pyarrow libraries, and this turned out to be prohibitively
difficult. I did get it to work on my laptop, but (i) it didn&39;t work on
the cluster due to CXX11 ABI mismatches and (ii) it made it impossible
to distribute binary wheels.

Passing atom data with numpy arrays does often require copying. The
reason is that the initial filtering step often causes the atoms
dataframe to be split into multiple chunks. I assume this is a
side-effect of some sort of parallel processing. Even so, the numpy
approach should still be better than having the whole loop in python,
which is what I had before. For one thing, this should cause many fewer
heap allocations. ([`6ac2f3e`](https://github.com/kalekundert/macromol_voxelize/commit/6ac2f3e32179148cca80a2f723a03138f000b8c8))

* feat: pass atoms dataframe to C++ via arrow

- This commit also replaces the regular-expression based channel
assignment algorithm with a much more efficient one. ([`3642a2a`](https://github.com/kalekundert/macromol_voxelize/commit/3642a2a2cee2e09b92032ca49b2319a42476014e))

Chore

* chore: remove changelog ([`6a926eb`](https://github.com/kalekundert/macromol_voxelize/commit/6a926ebce4cf5837943fa0c34e2aeaab4e4e44d1))

Feature

* feat: add option to render only outline/axes, not image ([`b0ab7c6`](https://github.com/kalekundert/macromol_voxelize/commit/b0ab7c6eb7299d03fbe8097e755628b6621ad174))

* feat: allow custom output dtypes

Other minor changes:

- Add some documentation.
- Provide a way to assign channels only to those atoms that pass the
initial filtering step. This is a slight performance optimization,
since assigning channels usually involves regular expressions, so can
be a bit slow.
- Make the occupancy column optional. ([`ea52659`](https://github.com/kalekundert/macromol_voxelize/commit/ea52659047e6a918cd567451f2ecc5525d74ae4e))

Fix

* fix: check that the atom radius parameter is a number ([`ccccbec`](https://github.com/kalekundert/macromol_voxelize/commit/ccccbecf9e9626ce073c1dbcd3fb91dce796ef40))

Test

* test: avoid testing on float128; not available on Mac ([`8984fe5`](https://github.com/kalekundert/macromol_voxelize/commit/8984fe56cfb99e33b7e736e2113af082d2741920))