Josiah
banner
josiahparry.com
Josiah
@josiahparry.com
oh, you know
im not sure what is involved in making extensions to torch in R but—this is working for me
github.com/mlverse/torc...
feat: GCN layer · Issue #1376 · mlverse/torch
I am working through a paper that uses generalized graph convolution (GCN) layers. Since there is no tooling in torch to do this presently, I created the layer itself. The implementation is below. ...
github.com
November 5, 2025 at 7:59 PM
rlang::list2(...)
October 18, 2025 at 7:08 PM
<3
October 10, 2025 at 7:32 PM
heard, thanks! :)
December 10, 2024 at 8:49 PM
you can use the @posit.co Ark kernel with Zed but it's not the console type vibe i want :/
zed.dev/docs/repl#in...
REPL - Zed
zed.dev
December 8, 2024 at 7:39 PM
At my last use, duckplyr uses the in-memory duckdb database which doesn’t support out-of-core processing. does it now set up a temporary duckdb database?
December 7, 2024 at 4:59 AM
say it with me!!!

duck👏🏼 plyr 👏🏼 doesn't 👏🏼 support 👏🏼 larger 👏🏼 than 👏🏼 memory 👏🏼 workloads 👏🏼 but 👏🏼 DuckDB 👏🏼 with 👏🏼 {dbplyr} 👏🏼 does
December 6, 2024 at 11:44 PM
We did this with Azure Blobfuse. This requires that you also have an .Rprofile to update your `.libPaths()` as well.

Just pure pain.
December 6, 2024 at 9:03 PM
databricks clusters are also ephemeral so if you want any libraries you have to install em with a startup script. One way to get around it is to use Docker container based images which have the packages you use installed. OR you use a startup script to mount the persistent dir to your runtime.
December 6, 2024 at 9:03 PM
What we did in my previous role was that we had 2 docker images: one for interactive development that had RStudio Workbench (now Posit Workbench) and another that did not have it. The latter was used for integration into the Databricks jobs—b/c you cannot use Job scheduler when using brickster
December 6, 2024 at 8:56 PM
(sorry, jeremy gonna give an outside view on this one :P )
the R + Databricks "story" tends to be "use your own infrastructure with an Databricks connect and use {dbplyr}". Like 90% of the use case of Databricks is their compute so having to standup additional infra is a tough call.
December 6, 2024 at 8:56 PM
its absolute crap.
December 6, 2024 at 8:53 PM
Awesome! there's also an example "highly performant" shiny app in there too
github.com/flrsh-dev/fl...
flrsh-lessons/duckdb-deep-dive/project at main · flrsh-dev/flrsh-lessons
{flrsh} lessons. Contribute to flrsh-dev/flrsh-lessons development by creating an account on GitHub.
github.com
December 6, 2024 at 8:33 PM
🦆 DB Building Blocks
flrsh.dev
December 6, 2024 at 6:42 PM
i started drafting what i thought was the most important about duckdb a while ago. never finished tho...classic.
github.com/flrsh-dev/fl...
flrsh-lessons/duckdb-deep-dive at main · flrsh-dev/flrsh-lessons
{flrsh} lessons. Contribute to flrsh-dev/flrsh-lessons development by creating an account on GitHub.
github.com
December 6, 2024 at 6:38 PM
multiple
December 6, 2024 at 6:25 PM
I cannot for the life of me figure out how to use this feature. It _seems_ so cool but I just dont get it
December 6, 2024 at 4:41 PM
timely! y'all mind sharing this job post? my team is hiring a dev :)

bsky.app/profile/josi...
December 5, 2024 at 8:04 PM