$ wget https://zenodo.org/record/162171/files/bayesian_inference_fbm.ap $ python -m venv venv $ source venv/bin/activate # Ubuntu $ source venv/Scripts/activate # Windows (venv)$ pip install numpy matplotlib h5py tempdir activepapers.py [ output omitted ] (venv)$ aptool ls code/generate_report code/inference_convergence code/lipid_analysis_l=10 code/lipid_analysis_l=100 code/lipid_analysis_l=200 code/lipid_analysis_l=50 code/python-packages/fbm ...
After this, I effectively got stuck.
The extraction worked, but resulted in a number of skips:
$ aptool checkout Skipping /code/python-packages/literate_python: data type reference not extractable Skipping /code/python-packages/unit_tests: data type reference not extractable Skipping /data/long_time_trajectory: data type reference not extractable Skipping /data/parameters/alpha_grid: data type data not extractable Skipping /data/parameters/alpha_in: data type data not extractable Skipping /data/parameters/n_traj_convergence: data type data not extractable Skipping /data/parameters/n_traj_ml_estimate: data type data not extractable Skipping /data/parameters/trajectory_lengths: data type data not extractable Skipping /data/short_time_trajectory: data type reference not extractable
On Ubuntu 20.04 (in WSL), I was able to update some parameters in
code/set_parameters.py, and have it do something:
$ aptool checkout code $ vim code/set_parameters.py # make changes $ aptool checkin code $ aptool update -v Dataset /data/parameters/alpha_in is stale or dummy, running /code/set_parameters Dataset /documentation/inference_convergence/convergence_fbm_l=100.pdf is stale or dummy, running /code/inference_convergence Dataset /documentation/short_time_modification/subsampling_convergence_l=100.txt is stale or dummy, running /code/short_time_modification
I don't know what this changed/updated and was unable to successfully perform any other
I am very familiar with Python in general, but had never heard of 'ActivePapers.py' before
Only had to
pip install ActivePapers.py + its requirements, no further software needed (apart from Windows Subsystem for Linux)
On Windows, all attempts at
aptool run <script> failed with import errors. It seems the ActivePapers tool runs into problems with importing the local module files from the
code/python-packages folder. This cost me a lot of time in trying workarounds before finally switching to Ubuntu (wsl).
unit_tests files in
code/python-packages/ seem to be missing. As noted earlier, they are skipped during checkout and only show up as empty files. This caused an error when trying to
aptool run any of the
$ aptool run lipid_analysis_l=10 Traceback (most recent call last): File "<bayesian_inference_fbm.ap>:/code/lipid_analysis_l=10", line 5, in <module> ... File "<bayesian_inference_fbm.ap>:/code/python-packages/gaussian_processes", line 12, in <module> ModuleNotFoundError: No module named 'unit_tests'
I really like the idea of having this kind of all-in-one file with logs, reproducibility and dependency checks built-in!
aptool commands are not difficult and are quick to learn.
My main suggestion would be better documentation. At least a local README along the
.ap file in Zenodo with basic instructions. For example:
Finally, the documentation of the ActivePapers tool itself should also be improved, but since it seems to be a (partially) abandoned project, I understand that is unlikely to happen.
Documentation could be better:
Installation instructions on this website were incomplete:
matplotlib was not listed as a requirement, but was needed to run any
Understanding how to use ActivePapers was difficult: only available documentation was a tutorial https://activepapers.github.io/python-edition/tutorial.html and the
aptool --help output.
Separate installation instructions and a basic tutorial using a public example were actually available, pleasantly written and were indeed helpful in getting started.
The previously suggested addition of a README, and checking the
.ap file for the missing code files.
I agree with the Author that it is an interesting test case in longevity. I feel it got very close to being an easily reproducible paper, despite the limited documentation and age of the used ActivePapers tool.