Some packages don't install through conda #63
Comments
|
FWIW, it could be possible to detect if the system is Arm (see flags in jlmelville/uwot#100) and have |
|
If this is something I can help with I'm happy to do so. Would it be useful if I went for a fresh manual env creation + install and reported here the versions installed by |
|
I am facing the same issue using the latest version of velociraptor (v1.10.0, Bioc 3.17) on an M1 macbook pro
Yes definitely... Wouldn't it also be a good opportunity to update Let me know if I can help! |
|
Woops. Thanks for the fresh comment. Completely fell off my radar (@LTLA 's too, I presume). I would also take the offer for a conda environment manifest. I'm swamped at the moment. |
|
I think we may have to upgrade |
|
I am experiencing a similar issue to @gdagstn and @julien-roux using an M2 Mac: Similarly, manually installing the packages works, however, only when the versions are not specified, e.g. But not: Conda output: Subsequently, SessionInfo> sessionInfo()R version 4.4.0 RC (2024-04-16 r86457) Matrix products: default locale: time zone: Europe/London attached base packages: other attached packages: loaded via a namespace (and not attached): |
|
Looking into it today. I've used the versions listed in #63 (comment) to try quickly update the versions of dependencies, but ran into I've also tried to figure out new environment specs a few different ways, all leading to various kinds of errors. Will continue investigating as time allows. Help welcome if anyone is keen. Can't blame anyone for the complexity of the task! |
|
Following mamba create -c conda-forge -c bioconda -n scvelo_0.2.5 scvelo=0.2.5
mamba activate scvelo_0.2.5
conda activate scvelo_0.2.5I get the environment But the error |
|
Note to self With: I successfully ran |
|
Good news is I'm making progress with scvelo actually running. Bad news is I'm running into theislab/scvelo#1212 |
😭 |
kevinrue
added a commit
that referenced
this issue
May 22, 2024
|
success b9717ca using # scvelo 0.2.5
# and dependencies pinned to fix various issues linked below
.scvelo_dependencies <- c(
"scvelo=0.2.5",
"matplotlib=3.7.3", # https://stackoverflow.com/questions/77128061/ydata-profiling-profilereport-attributeerror-module-matplotlib-cbook-has-no
"tqdm=4.66.4", # required for progress bar
"ipywidgets=8.1.2", # required for progress bar
"jupyterlab=4.2.0", # required for progress bar
"numpy=1.23.1" # https://github.com/OpenTalker/video-retalking/issues/35
)now the tedious task of pinning all the other dependencies to make my current M1 environment 100% reproducible on other M1's |
Merged
|
Anyone willing to try out the updated versions on branch |
|
I tried it on an M1 - the |
|
indeed this is my side by side comparison (left: branch kra-63 v 1.15.1; right: latest github vignette v1.9.2) 
seems to me it's more an issue with the t-SNE that looks more 'scattered' potentially forcing the vectors to stretch further (I don't know how to formally check if that's the case) |
|
for a more up to date comparison, on the right, v1.14.0 (current bioc release) 
|
|
problem solved in 975bd9e essentially, the move from scvelo to scanpy for neighbor calculation changed the default value of n_neighbors from 30 to 15 rather than forcing the old default value, the vignette documents the change and educates users to explicitly set the value to 30 themselves. (new default is thus 15) |
|
For the record, after fix, the t-SNE looks like this
|
kevinrue
added a commit
that referenced
this issue
May 30, 2024
* bump x.y.z version to even y prior to creation of RELEASE_3_19 branch * bump x.y.z version to odd y following creation of RELEASE_3_19 branch * #63 (comment) * success! * update comment * exact match * pin all package versions * Revert "pin all package versions" This reverts commit 75035b2. * update scvelo to 0.2.5 * propose dynamic selection of scvelo dependencies * add dependencies required for the same reason as M1 * add linux environment * fix man page * add Linux environment * bump bioc-devel version * update scvelo to 0.3.2 on M1 * update scvelo to 0.3.2 on linux * pin environment for windows * support Linux, not LinuxAarch64 * remove outdated comment * docs, version bump and news * use old value of 30 neighbors * add news --------- Co-authored-by: J Wokaty <[email protected]> Co-authored-by: Kevin Rue-Albrecht <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi,
I am working on a M1 Macbook Pro and I had some issues running
scvelo()(from the Bioconductor 3.16 release) due to some packages being apparently unavailable:Since the proper arm64 miniconda version was installed, I was able to circumvent this by running the creation of the conda environment and installation of the
scvelomodule manually, i.e.It works properly after this, but I do see that this is likely going to install slightly different package versions from the ones specified in the original manifest in velociraptor, so I'm not sure this hacky fix is the way to go.
Is there something else I should do/should have done?
sessionInfo()
The text was updated successfully, but these errors were encountered: