Init: custom docker container for Data Science

I have enabled the TCP on port 8787 via UFW, and am running a docker container of the image ‘rocker/tidyverse’, which is providing an Rstudio IDE online to play with.

This was relatively straightforward, starting with installing docker on my debian machine, and then pulling in the rocker/tidyverse container, as well as the rocker/shiny-verse container. The plan is to host the shiny apps that I built on my VPS via the shiny server.

Though docker can be viewed as a dependency, the point is that deployment has been relatively easy. There appears to be some level of control. For example, I can deploy a container on 8788, but cannot access it until the port is enabled on UFW.

  • [ ] During my search, I did find some reports of vulnerabilities with using Docker on VPS, and being able to bypass UFW.

Since my earlier notes on docker included installation snippets and notes in org mode files, I could copy portions of the code into a temp org file on the server (in the Emacsclient), and run most of the commands from within Emacs, and Eshell. Though a proper terminal has it’s advantages as is anyway available at a keypress in tmux – eshell allows an easier method to search through the output.

So far, running some basic plots on the browser based Rstduio – only about 450MB of RAM has been utilised. This includes running tmux, within which run an Emacs daemon + client, weechat (IRC + Slack) and LAMP server, which is serving up the website / blog.

As long as the container does not become too heavy, I guess I would like umap, parsnip and other machine learning libraries already installed via docker. Considering how fast it was to get started with tidyverse – it may be fast enough, even with some additional libraries.

Installing umap

  • Note taken on [2019-08-12 Mon 11:37]
    The issue was with the swap. On a second attempt with a fresh container :- there were no problems in installing UMAP in a single shot.

umap would not install on the tidyverse container on the first go. The package Rspectra needed to be installed manually.

Error excerpt :

/usr/local/lib/R/etc/Makeconf:176: recipe for target 'fastLm.o' failed
Warning in install.packages :
  installation of package ‘RcppEigen’ had non-zero exit status
Warning in install.packages :
  installation of package ‘RSpectra’ had non-zero exit status
Warning in install.packages :
  installation of package ‘umap’ had non-zero exit status

Leave a Reply

Your email address will not be published. Required fields are marked *