v0.9.0

✨ Highlights

• This PR implements a new solver, which is now used by default (#625,
  @michaeldeistler). The new solver has the following advantages:
  • Much lower runtime. Across several morphologies, we get a 20% runtime speedup on
    CPU and a 50% runtime speedup on GPU.
  • Almost zero compile time. We get a 50x compile time speedup on CPU and a 3x compile
    time speedup on GPU.
• Utility to delete parts of a morphology (#612, @michaeldeistler):

from jaxley.morphology import morph_delete
cell = morph_delete(cell.branch([1, 2]))

• Utility to connect two cells into a single cell (#612, @michaeldeistler):

from jaxley.morphology import morph_connect
cell = morph_connect(cell1.branch(1).loc(0.0), cell2.branch(2).loc(1.0))

🧩 New features

• the default SWC reader has changed. To use the previous SWC reader, run
  jx.read_swc(..., backend="custom"). However, note that we will remove this reader
  in the future. If the new SWC reader is causing issues for you, please open an issue
  (#612, @michaeldeistler)
• radiuses are now integrated across SWC coordinates, not interpolated
  (#612, @michaeldeistler)
• remove pin of JAX version. New JAX versions (JAX>=0.6.0) resolve slow CPU
  runtime, see here (#623, @michaeldeistler)
• running the d_lambda rule is now much faster, see
  the how-to guide
  (#625, @michaeldeistler)

📚 Documentation

• Introduce the how-to guide on the website (#612, @michaeldeistler)
• reorganize the advanced tutorials into subgroups (#612, @michaeldeistler)
• split the morphology handling tutorials into two notebooks (#612, @michaeldeistler)

🛠️ Internal updates

• improvements to graph-backend for more flexibility in modifying morphologies (#613,
  @michaeldeistler)
• remove root compartment for SWC files (#613, @michaeldeistler)
• enable traversing compartmentalized graph for optimizing solve order (#613,
  @michaeldeistler)
• ._comp_edges are being tracked in the View (#621, @michaeldeistler)
• introduce ._branchpoints_ attribute and track in the View (#612, @michaeldeistler)
• pre_global_comp_index and post_global_comp_index are no longer part of .edges
  by default. To get them, do net.copy_node_property_to_edges("global_comp_index")
  (#625, @michaeldeistler)

🐛 Bug fixes

• ChainTransform forward now working as mirror to inverse (#628, @kyralianaka)
• allow data_set with vectors of values (#606, @chaseking)

🎉 New Contributors

• @chaseking made their first contribution in #606