search

qupath / qupath-docs

QuPath documentation
http://qupath.readthedocs.io/
17 stars 22 forks source link

Slight inconsistency in GPU docs #100

Closed alanocallaghan closed 10 months ago

alanocallaghan commented 10 months ago

I would suggest we maybe just talk about conda here, because introducing mamba as an alternative is not necessary overhead, despite the fact we're saving people a bit of time in setting up their envs.

petebankhead commented 10 months ago

Should the download link then point to conda?

I've no strong opinions on it, I also thought sticking with conda might be less confusing for people since it's what is included on the PyTorch website.

I admit I was a bit influenced by @haesleinhuepf's article (and the fact I also tend to find conda much slower).

alanocallaghan commented 10 months ago

I definitely agree people should use mamba - some of the environments I've built take like 30 minutes with the standard resolver. I just think it's not relevant to the problem here - maybe better for a footnote

petebankhead commented 10 months ago

Ah ok, I misread the PR in a rush. Merged to fix the inconsistency for now. Changes to shorten and simplify the instructions welcome - can see the argument for just embracing conda here, as long as we check it actually works (since mamba defaults to conda-forge, as far as I recall, I'm not sure if there are any other hidden gotchas - I only tested the instructions with mamba as I wrote them).

haesleinhuepf commented 10 months ago

(since mamba defaults to conda-forge

mamba-forge does that. But the project was discontinued.

I think you folks can safely mention conda only as mamba will become the new solver within conda. If it didn't happen yet, it's just a matter of time.

petebankhead commented 10 months ago

Excellent, I hadn't heard of that at all, thanks @haesleinhuepf!