Skip to content

running.md

Latest commit

 

History

History
47 lines (34 loc) · 1.75 KB

running.md

File metadata and controls

47 lines (34 loc) · 1.75 KB

Running existing images

Assuming Bob has a file named tree.py. Assuming this piece of code accepts command line parameters specifying data and target space, performs three fold cross validation and returns the result, so something like:

python tree.py --super_big_data somedata.dat --super_multitarget_space targets.dat

and returns something in the lines of:

RESULT10fCV 1 0.89
RESULT10fCV 2 0.85
RESULT10fCV 3 0.81

Of course, Bob is interested in obtaining the above results of cross validation for some new data. Assuming that Bob is not able to exactly replicate the environment on his own for one of the following reasons:

  1. Some driver conflict prohibits installation of one of the libraries
  2. He's unable to install all requirements.txt directly for whatever reason
  3. He's lazy

We are able to present to him a simple solution - the Singularity container. The idea is the following:

  1. Precompile a Python environment with minimal spatial overhead and all required libraries
  2. Share the environment as a single file

What are singularity files?

Singularity files are images, just like e.g., your favourite Star Wars movie DVD. Given an example Singularity image, e.g., OurSingularityImage.sif, the above piece of code can be run as:

singularity exec OurSingularityImage.sif python tree.py --super_big_data Bob_somedata.dat --super_multitarget_space Bob_targets.dat
  • singularity = call of the tool
  • exec = flag to execute a given command (there is also run)
  • OurSingularityImage.sif = our environment image.

and the output should be, of course, different. E.g.,

RESULT10fCV 1 0.31
RESULT10fCV 2 0.32
RESULT10fCV 3 0.33

That's it. Bob was able to replicate our environment and run the codebase.