evalr-experiment

This is a skeleton project for running evalr experiments.

Getting started guide

$ git clone ssh://git@github.com/PRL-PRG/evalr-experiment
$ cd evalr-experiment

Important: all of the following commands should be run inside the cloned repository!

If the docker image has not yet been created, run

$ make -C docker-image

Get into docker

$ make shell

Get the dependencies. The reason why we do not put them yet in the image is that we might want to have some local changes to them.

docker% make libs-dependencies

Install libraries

docker% make libs

Install packages from packages.txt. This has to be done every time, this file changes.

docker% make install-packages

The pipeline

Tracing CRAN packages

To trace packages, run:

docker% make package-trace-eval

The result should be always in in run/package-trace-eval. Older runs will be renamed to run/package-trace-eval.<X>.

To preprocess the data, run:

docker% make package-preprocess

The result should be always in in run/preprocess/package. Older runs will be renamed to run/preprocess/package.<X>. The preprocess will always use the latest data from the package-trace-eval task. It will not trigger a new tracing.

Tracing core packages

TBA

Tracing Kaggle

TBA

Debugging when something is wrong

TBA

Debugging using gdb

The image has both gdb and gdbgui installed. To start a gdbgui session do:

$ make shell
docker% R -d 'gdbgui -r'
Warning: authentication is recommended when serving on a publicly accessible IP address. See gdbgui --help.
View gdbgui at http://172.17.0.3:5000
View gdbgui dashboard at http://172.17.0.3:5000/dashboard
exit gdbgui by pressing CTRL+C

and then connect to the IP address it suggested.