Feature request: Ability to run BAM parsing, fitting and clustering independently

[shiraz-shah]

Dear VAMB team,
In our HPC setup we have separate servers optimsed for storage access, compute and GPU compute. Therefore it would be useful for us to be able parse the BAM and fasta files on one server (the one with fast access to our fastq long term storage), then copy over the composition and abundance.npz files to the GPU workstation for fitting and clustering.

Likewise it would be useful to be able to run clustering separately after copying latent.npz from the GPU workstation to the CPU server (as we have experienced problems with clustering on GPU, as noted earlier). This would also allow for straightforward reclustering of multiple dumped latent representations as discussed earlier.

Obviously not urgent, but highly appreciated!

[jakobnissen]

That's a good idea. When we've done that internally, we've used scripts that reach into Vamb internals, but as Vamb grows more complex, that's getting harder and harder to get right.