2024 Reproducible work in Data Science (X. de Pedro)

"Data Science. Applications to Biology and Medicine with Python and R", at IL3 - University of Barcelona. April 10th, 2024 (16-19:15h).

Content at https://seeds4c.org/reproduciblework2024

Start Slideshow Presentation Slides in PDF

1. Introduction - the problems (i)

	-->
Obsolete Devices storing code & data	-->	Ease copying to new devices (legally also: copyleft, ...) + online repositories

1.1. The problems (ii)

	-->
Software obsolescence and incompatible dependency versions	-->	Adapt to code evolution: - Controlling Package Versions ( `renv` ) - VCS (git, bazaar, svn...) VCS = Version Control Systems

1.2. The problems (iii)

	-->
Centralization (such as Subversion VCS (svn) may increase efficiency but it also decreases Resilience ("shit happens")	-->	From Centralized VCS (such as svn) to Decentralized VCS (such as git) VCS = Version Control Systems

1.3. The problems (iv)

	-->
Conflicting package versions at system level with package versions at project levels	-->	Package versions per project Environment ( `renv` )

1.4. The problem (v)

Sometimes a project was developed with a major version of a programing language (R 3.x, Python 2.x), while another project in the same server requires a different major version (R 4.x, Python 3.x)	-->	R case: from RStudio Server to Posit Workbench (former RStudio Server Pro) You can choose R version per project Python: Several approaches (conda, PyCharm, ...): see this as an example.

1.5. The problem (vi)

	-->
Changes per Operating System itself (32 bit systems unsuported anymore, discontinued linux distros, ...)	-->	Virtual Machines or Containers (OVA, KVM, LXC, Docker, Pod...): you can choose OS version per container within the same server

2. Enemies of reproducibility & adaptability

Enemies of reproducibility and adaptability (in levels): Changes / Evolution / Versions!

Operating system and its dependencies (and their versions)
Programming language (and its version)
Specífic Packages (and versions) as dependencies for your Work Project
Versions of your own code (algorythm and param variations, etc): lacking versioning system
Readability and tidyness of your own code / routines / scripts
Lack of documentation/help resources + steep learning curve to use it or adapt it to your context or infrastructure

3. Reproducibility & Adaptability

How to avoid reproducibility & adaptability enemies (in R & Python for Data Science):

ISSUES	SOLUTIONS / WORKAROUNDS
(Level 1) Versions in OS repos & critical dependencies: curl, ssl, GDAL, Java, cpp, V8...	Virtual Machines or Containers (VBox, KVM, LXC, Docker, Pod...)
(Level 2) Versions in Programming language: Python 2.x vs 3.x, R 3.x vs 4.x, ...	Python: Conda, ~~Google Colab~~, ... R: RStudio/Posit Workbench General (in Linux clusters): software modules.
(Level 3) Versions in Specific packages	=== Py: .env, poetry R: ~~Packrat~~, Renv (by versions), MRAN (by date)
(Level 4) Versions in Your own scripts	Decentralized VCS: Git (Gitlab, Github, ...), Bazaar (Launchpad), ... Centralized VCS: CVS, SVN (Sourceforge, ...), ... VCS = Version Control system
(Level 5) Tidy script content and organization	Literate Coding (Scripting & Coding) / Analysis - R: Rstudio Notebooks with modern R (Tidyverse). VS Notebooks, G-Colab, ... - Python: Jupyter Notebooks, Rstudio Notebooks, VS Notebooks, G-Colab, ... ( `Quarto` Mardown and rendering for both and more)
(Level 6) Help to lower the learning curve	Documentation, Code Vignettes, Examples, Tutorials, Learning material ( `learnr` ), Books ( `bookdown` )...

4. Reproducibility & Adaptability - Example in Posit Cloud

Example in https://posit.cloud (former RStudio Server Pro) :

Level 1: A Container with a specific linux distro (e.g. Ubuntu Linux 20.04 Focal LTS) per project.
Level 2: RStudio/Posit Workbench (which allows choosing R version per project)
Level 3: renv for your R package collection (and specific versions) in your project
Level 4: git or svn for your scripts in your project
Level 5: YOU (Tidyverse is your friend)
Level 6: YOU (+ helpers: roxygen2, blogdown, learnr, bookdown, ...)

4.1. Level 1: Virtual Machines or Containers

From:
https://kubernetes.io/docs/concepts/overview/

4.2. Level 2: RStudio-Posit Workbench

4.3. Level 3: renv - for packages

Version control in work "environments"

4.3.1. Virtual environments in R with renv

4.3.2. From utils: : sessionInfo() to renv: : snapshot() + renv.lockalso fails

utils::sessionInfo()

> sessionInfo()
R version 4.1.2 (2021-11-01)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 22.04.1 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.10.0
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0

locale:
 [1] LC_CTYPE=ca_ES.UTF-8       LC_NUMERIC=C               LC_TIME=ca_ES.UTF-8       
 [4] LC_COLLATE=ca_ES.UTF-8     LC_MONETARY=ca_ES.UTF-8    LC_MESSAGES=ca_ES.UTF-8   
 [7] LC_PAPER=ca_ES.UTF-8       LC_NAME=C                  LC_ADDRESS=C              
[10] LC_TELEPHONE=C             LC_MEASUREMENT=ca_ES.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices datasets  utils     methods   base     

other attached packages:
 [1] kableExtra_1.3.4 fs_1.5.2         tictoc_1.1       lubridate_1.9.0  timechange_0.1.1
 [6] janitor_2.1.0    knitr_1.40       markdown_1.3     RODBC_1.3-19     fst_0.9.8       
[11] forcats_0.5.2    stringr_1.4.1    dplyr_1.0.10     purrr_0.3.5      readr_2.1.3     
[16] tidyr_1.2.1      tibble_3.1.8     ggplot2_3.4.0    tidyverse_1.3.1 

loaded via a namespace (and not attached):
 [1] httr_1.4.4        jsonlite_1.8.3    viridisLite_0.4.1 modelr_0.1.10     assertthat_0.2.1 
 [6] renv_0.16.0       cellranger_1.1.0  yaml_2.3.6        pillar_1.8.1      backports_1.4.1  
[11] glue_1.6.2        digest_0.6.30     rvest_1.0.3       snakecase_0.11.0  colorspace_2.0-3 
[16] htmltools_0.5.3   pkgconfig_2.0.3   broom_1.0.1       haven_2.5.1       scales_1.2.1     
[21] webshot_0.5.4     svglite_2.1.0     openxlsx_4.2.5.1  rio_0.5.29        tzdb_0.3.0       
[26] generics_0.1.3    ellipsis_0.3.2    withr_2.5.0       cli_3.4.1         magrittr_2.0.3   
[31] crayon_1.5.2      readxl_1.4.1      evaluate_0.18     fansi_1.0.3       xml2_1.3.3       
[36] foreign_0.8-82    tools_4.1.2       data.table_1.14.4 hms_1.1.2         lifecycle_1.0.3  
[41] munsell_0.5.0     reprex_2.0.2      zip_2.2.2         compiler_4.1.2    systemfonts_1.0.4
[46] rlang_1.0.6       grid_4.1.2        fstcore_0.9.12    rstudioapi_0.14   rmarkdown_2.18   
[51] gtable_0.3.1      DBI_1.1.3         curl_4.3.3        R6_2.5.1          fastmap_1.1.0    
[56] utf8_1.2.2        stringi_1.7.8     parallel_4.1.2    Rcpp_1.0.9        vctrs_0.5.0      
[61] dbplyr_2.2.1      tidyselect_1.2.0  xfun_0.34        

>

renv::snapshot() i ./renv.lock

{
  "R": {
    "Version": "4.1.2",
    "Repositories": [
      {
        "Name": "CRAN",
        "URL": "https://cloud.r-project.org"
      }
    ]
  },
  "Packages": {
    "DBI": {
      "Package": "DBI",
      "Version": "1.1.3",
      "Source": "Repository",
      "Repository": "CRAN",
      "Hash": "b2866e62bab9378c3cc9476a1954226b",
      "Requirements": []
    },
    "tinytex": {
      "Package": "tinytex",
      "Version": "0.42",
      "Source": "Repository",
      "Repository": "CRAN",
      "Hash": "7629c6c1540835d5248e6e7df265fa74",
      "Requirements": [
        "xfun"
      ]
    },
    "tzdb": {
      "Package": "tzdb",
      "Version": "0.3.0",
      "Source": "Repository",
      "Repository": "CRAN",
      "Hash": "b2e1cbce7c903eaf23ec05c58e59fb5e",
      "Requirements": [
        "cpp11"
      ]
    },
    "zip": {
      "Package": "zip",
      "Version": "2.2.2",
      "Source": "Repository",
      "Repository": "CRAN",
      "Hash": "c42bfcec3fa6a0cce17ce1f8bc684f88",
      "Requirements": []
    }
  }
}

4.3.3. "Happy path"

For a reproducible environment

Commands in terminal - Computer 1

cd project_folder
git init
R
[obrir projecte de RStudio]
renv::init() # to initialize renv in your code project
renv::snapshot() # to make a snapshot "picture" of the list of R packages used within the whole R project and their respective package versions
q()
git commit ...
git push

Commands in terminal - Computer 2

cd project_folder
git clone/git pull ...
R
[open same RStudio project]
renv::status() # for a report on which steps are suggested for you to follow
renv::restore() # to restore the package library (with the required pacckage versions) for this project
[continue working in/developing your code]
renv::snapshot() # to make a new snapshot "picture" (in case there are new packages and/or versions or R packages neweer or older in use in your project ;-) )
q()
git commit ...
git push

4.3.4. Infraestructure

Projects with renv write and use these files in order to work:

File	Use
.Rprofile	Used to activate renv for new R sessions launched in the project.
renv.lock	The lockfile, describing the state of your project’s library at some point in time.
renv/activate.R	The activation script run by the project .Rprofile.
renv/library	The private project library.
renv/settings.dcf	Project settings – see `?settings` for more details.

By default, renv uses a package memory-cache here:

Platform	Location
Linux	`~/.local/share/renv`
macOS	`~/Library/Application Support/renv`
Windows	`%LOCALAPPDATA%/renv`

4.3.5. Advanced use

renv::install("packagename", version="0.1") # to install old versions from a package (useful also for discontinued packages in CRAN!). See possible package-version numbers at https://cran.r-project.org/src/contrib/Archive/yourpackage/
renv::record("packagename", version="0.1") # to save at renv.lock the specific version you need for this package
renv::deactivate() # to temporarily deactivate renv in your project
renv::activate() # to reactivate renv in your project
renv::equip() # for special installations in MS Windows
vignette("docker", package = "renv") # for a commbined use with Docker
vignette("collaborating", package = "renv") # to improve collaborative use in work teams

And much more. See:

4.4. Level 4: git - for code

See: https://gitlab.com/radup/curs-r-introduccio/ > Folder "codi" > 10.compartir.via.git.Rmd (or .pdf)
See also my own git recipes over some years, github cheatsheet, ...: https://seeds4c.org/git

5. More information

Work Environments in R

https://solutions.posit.co/envs-pkgs/environments/

Videos

An Introduction to Reproducible Research Practices. 29 d’abr. 2022. John Little. Duke University. Video
Designing a Reproducible Workflow with R and GitHub. John Little. 22 de nov. 2021 Video | Tutorial
The workflowr R package: a framework for reproducible and collaborative data science. 13 de jul. 2018. R Consortium. Video
Kevin Ushey | renv: Project Environments for R | RStudio (2020). Posit PBC.. 20 de des. 2020. Video

R Packages

renv | workflowr | learnr | roxygen2 | Tidyverse

Free Work environments for Collaborative Data Science with R & Python

https://posit.cloud/plans/free

Additional tutorial with big data to follow on site (R Cloud)

Danielle Navarro. 2022. “Using Amazon S3 with R” March 17, 2022.

Papers

Wallach JD, Boyack KW, Ioannidis JPA. (2018) Reproducible research practices, transparency, and open access data in the biomedical literature, 2015–2017. PLoS Biol 16 (11): e2006930. https://doi.org/10.1371/journal.pbio.2006930
Leek JT, Peng RD. Opinion: Reproducible research can still be wrong: adopting a prevention approach. Proc Natl Acad Sci U S A. 2015 Feb 10;112(6):1645-6. doi: 10.1073/pnas.1421412111. PMID: 25670866; PMCID: PMC4330755

6. Hands-on practical exercise

6.1. Register a free account at Posit Cloud

You can do so at:

https://posit.cloud/plans/free

You will need to click on a link sent to your email inbox to validate your account.

Once done, you'll see something like:

6.2. Create a Project from git repository

Enter Posit cloud and click at New Project > New Project from Git Repository

6.2.1. Visit gitlab to get clone url

Visit this code project in gitlab to get the project clone url:
https://gitlab.com/xavidp/datascience2023

6.2.2. Create project from git repo

Paste it in the Posit cloud popup window and click at OK:

6.3. Choose R 3.6.x & Run Rmd

6.3.1. Install dependencies also

6.3.2. Running Rmd will perform GNU/Linux system commands also

GNU/Linux system commands will usually be much more efficient in memory & cpu

It helps to prevent RAM bottlenecks with just 1Gb RAM on Posit Cloud Free plan

(while csv file from reduced meteorological dataset is already 0.5 Gb).

6.3.3. Display raw data

Variables are in numeric codes (not easily readable by humans in a semantic way). We lack some varaible names (or acronyms at least) for readability.

6.3.4. Transform in tidy way (i)

6.3.5. Transform in tidy way (ii) - result

6.3.6. Last code chunks

6.4. Choose R 4.2.x & Run Rmd again

Repeat the previous steps but in a R 4.2.x environment: install dependent R packages again... (new environment, but still installing from CRAN repos). renv not needed in this case still (lucky you!).

So far, so good.

6.5. Choose R 3.4.x & Run Rmd

Now let's touch some issues with R package versions in a R 3.4.x environment

Running Rmd will fail at some package installations

dplyr installation fails
readr is reported as unavailable in R 3.4.4
tidyr installation also fails (as well as purrr )

Solution

In this case, the solution involves finding some valid previous package version for each conflicting R package, and using this type of commands:

renv::init()
renv::install("packagename@x.y.z") # being x.y.z a valid package version number, as taken from https://cran.r-project.org/src/contrib/Archive/packagename/
renv::record("packagename@x.y.z")
renv::snapshot() # after all packages installed without any more issues

6.5.1. Use renv.lock recipe (i)

Let's get renv to the rescue. Once somebody solved these issues, and found a valid recipe of package versions for this environment, a file ./renv.lock will have been produced in the project root folder after running the command renv::snapshot()

I did this already, and I uploaded the produced renv.lock file to the manually created ./recipes/ folder in this project as a backup for you (as renv_R344.lock).

You can then copy now the ./recipes/renv_R344.lock file provided in the project as ./renv.lock in the project root folder, for renv to be able use it.

6.5.2. Use renv.lock recipe (ii)

Run renv::init() in the R console.

Choose restore the renv.lock package versions:
"1. Restore the project from the lockfile"

6.5.3. Use renv.lock recipe (iii)

You will be ready to go with minimum human intervention.

All R packages will be installed in the backgound to their required package versions, following the recipe that someone created for R 3.4.4. already.

The key file is the renv.lock file.

6.5.4. Use renv.lock recipe (iv) - finished

6.6. Additional info

Project (Container) goes to sleep on inactivity

Thanks

Xavier de Pedro Puente, Ph.D. - xavier.depedro@seeds4c.org

Slides available at:
https://seeds4c.org/reproduciblework2024

Unless elsewhere noted, contents of this web site are released under a Creative Commons license.

Please wait

Structures

Page actions