IMDReader removed from imdclient

[amruthesht]

Fixes #53

This PR assumes that IMDReader has been integrated into MDAnalysis - PR

Changes made in this Pull Request:

• IMDReader and StreamReaderBase has been removed from the main package (and potentially moved to MDAnalysis PR) i.e. files IMD.py and streambase.py
• Tests for the IMDReader have been removed from the tests folder (and potentially moved to mdanalysis testsuite)

PR Checklist

• Tests?
• Docs?
• CHANGELOG updated?
• Issue raised/referenced?

[amruthesht]

TO-DOs

Get rid of any usage or references to MDAnalysis (as suggested by @orbeckst)

• Remove any MDAnalysis usage for generic tests
• Make simple pseudo "reader" that can be used for tests

[orbeckst]

Remove MDAnalysis from pyproject.toml etc — there shouldn't be a any place in the package where it gets installed or imported.

For right now, streamanalysis is in a weird place. We can leave it in there and test it or we can disable/remove it. Something to discuss.

[orbeckst]

As discussed, we can use MDAnalysis for testing (reading trajectories) but not for streaming itself.

[orbeckst]

My primary issue with the current approach is that it still requires IMDReader for testing. The way that we had concluded our discussion (ASU+MDA meeting on 2025-03-03) was the decision

imdclient: allow MDA as test dependency for imdclient tests (but do not allow IMDReader)

In my comment below I am outlining how you can use MDAnalysis to construct a Universe after obtaining the data from the stream as numpy arrays. For this to work you need to write code that only uses imdclient to pull data from a stream.

[orbeckst]

I still don't think that u = mda.Universe((tmp_path / topol), f"imd://localhost:{port}") should be used because it uses IMDReader. I understand that it makes it easier to write integration tests. But it really muddles the water and really tests MDAnalysis and only indirectly imdclient. I believe it should be relatively straightforward to pull the data from the streame with imdclient only and then put it into a Universe (with in-memory representation) for further processing and comparisons.

You should be able to

1. pull the coordinates and velocities from the stream as numpy arrays

2. save them as a trajectory: create a Universe from the topology and the numpy array with the memory reader:

   u = mda.Universe(tmp_path / topo, coordinate_array, dt=dt)

   where coordinate_array is a numpy array of shape (n_frames, n_atoms, 3) and dt is the time between frames in ps.

   There is also a way to create a Universe with velocities and forces. One might have to use MemoryReader directly and associate it ... maybe just passing the memoryreader instance as a trajectory file will work.

3. compare the original trajectory and your new universe (with in-memory data).

[amruthesht]

Thank you @orbeckst for the detailed suggestion.
Unfortunately, the eaxct apporoach as above would be a little convoluted as all the inof the in the buffer is not readily available as a NumPy array. Instead the IMDCleint object reads one frame at a time. Thus, I'm implementing a minimalReader class that would essentially use IMDClient to read one frame at a time; and then compare that with the data in the files (read in as a MDA universe using MDA File Readers)

[ljwoods2]

Why not just iterate through the stream and copy each numpy array out of the IMDFrame and into an array like Oliver is describing?

[ljwoods2]

Right now we're doing almost the same thing except now we're wrapping stream iteration in a Universe and then writing it to trr instead of an in-memory numpy array