Recently the original virtualenv-baseed approach started to fall apart,
because mutual incompatibilities between the dependencies (some pinned
at old versions) increased.
This changeset provides a container recipe to generate a modern (or
most modern, given todays possibilities) environment to still
reproduce the analysis.
The container image can be built with
```
docker build -t remodnav:latest container
```
And the manuscript can be generated with
```
docker run -v $PWD:/tmp -w /tmp --rm --interactive --tty \
-u $(id -u):$(id -g) -e HOME=/tmp \
remodnav:latest \
make -f Docker-Makefile main.pdf
```
This does not reproduce everything exactly. There is a small (I'd say
insignificant) change in one panel of a Fig 4. The full backstory
and analysis of the cause is in
https://github.com/psychoinformatics-de/paper-remodnav/issues/20.
However, besides that **all** reported statistical scores and
performance measures reported in the paper remain identical.
Not that execution inside the container requires all inputs to be
provisioned before. Here is the specification
- remodnav/remodnav/tests/data/anderson_etal
- remodnav/remodnav/tests/data/studyforrest
- data/studyforrest-data-eyemovementlabels/*/*.tsv
- data/raw_eyegaze/sub-*/ses-movie/func/*_recording-eyegaze_physio.tsv.gz
- data/raw_eyegaze/sub-*/beh/*_recording-eyegaze_physio.tsv.gz
This will again be automated in a future changeset, based on a
container-run provenance record.
