{targetsboard}: a Shiny + React board for {targets} pipelines
This weekend’s project was {targetsboard}, an alternative Shiny + React UI to monitor and (perhaps one day) control {targets}’ DAGs. For those familiar with {targets}, it serves as an alternative to the default tar_watch().
I absolutely LOVE the {targets} package, and I mostly did this project for fun (and it was indeed fun!). However, it will take some time to make it minimally viable, so I’m publishing it with the hope of finding skilled folks to help take the idea and develop a proper version.
The idea
The main concept of this project is to propose a data orchestration tool designed with Data Scientists in mind, emphasizing reproducibility, reporting, and a robust feedback loop. It would be fantastic to have an Airflow-grade tool centered on Data Science!
And honestly, I believe that R has what it takes to be a leading force in this area. R has the potential to be an enterprise-ready, first-class citizen in MLOps. Packages like {targets}, {shiny}, {reactR}, {tidyverse}, {quarto}, {vetiver}, {sparklyr}, {dbplyr}, etc., are simply AMAZING!
How to try it
The package is at an ultra-early, experimental stage. Install the development version from GitHub:
remotes::install_github("Athospd/targetsboard")In your normal, day-to-day targets workflow, launch the board with tar_board(). It runs in the background and is served at localhost:9999:
library(targetsboard)
# launch the app
app <- tar_board()
# go to localhost:9999 where the board is being served
browseURL("http://localhost:9999")
# you can check it is alive, running on background
app$is_alive()It keeps track of your targets’ metadata just like tar_watch() would. Run your pipeline as usual and watch the DAG update:
tar_make()When you are done, shut the app down:
app$kill()Packages to put under your radar
- {targets}: amazing data orchestration with a focus on data science.
- {reactR}: a powerful templating toolkit to connect R with React components (I used the reactflow React lib as the DAG interface).
- {shiny}: now also available for Python, a rich web app framework, awesome for us mortal data scientists.
Let’s discuss it!
What orchestration tools do you know that are already DS-friendly? What data orchestration tool do you like the most? What are your thoughts on my beloved R in the MLOps field? Please let me know.