switch to Anaconda for installation

[Garrafao]

Instead of providing a requirements.txt, we should provide installation instructions for Python Anaconda because this will allow to use the graph-tool library.

[Garrafao]

This has been implemented with the last commit. @AinaIanemahy Could you please test the tool installation options described in the readme and to suggest changes if necessary?

[AinaIanemahy]

It seems that I still need mlrose and pyvis in the environment. Can we add mlrose to the pip dependencies in packages.yml lines 100 ff. I also need pyvis in this file.

- python-dateutil=2.8.2=pyhd3eb1b0_0
- pyvis=0.3.1=pyhd8ed1ab_0 # added this line
- qt-main=5.15.2=h53bd1ea_10

Also: Did you take #42 into account for the new environment? It seems like you still pip install mlrose in install_packages.sh. For me it only works with the git repo pip install git+https://github.com/gkhayes/mlrose. I'm not sure about installing git repos through packages.yml, but installing with pip in the activated repository works just fine.

[AinaIanemahy]

With respect to the graph tool I suggest doing the following:

• create a new env with the graph tool conda create --name wug -c conda-forge graph-tool
• then use the .yml as follows conda env update --file packages.yml

For DURel this might also work, but I still have to test.

[AinaIanemahy]

Also with respect to DURel: I need to make sure that wugs is activated. Therefore I need to execute the following lines at the beginning of run_system2.sh:

source /opt/conda/etc/profile.d/conda.sh
conda activate wug

I'm not sure whether we can add these lines without interfering with the execution of the pipeline outside DURel.

[Garrafao]

Also with respect to DURel: I need to make sure that wugs is activated. Therefore I need to execute the following lines at the beginning of run_system2.sh:

source /opt/conda/etc/profile.d/conda.sh
conda activate wug

I'm not sure whether we can add these lines without interfering with the execution of the pipeline outside DURel.

We can as this script is there exclusively for use in DURel. I suggest that you change the script according to your purposes. Please push changes immediately to this repo. Ideally, after doing changes, run test.sh to see whether the pipeline works.

[Garrafao]

It seems that I still need mlrose and pyvis in the environment. Can we add mlrose to the pip dependencies in packages.yml lines 100 ff. I also need pyvis in this file.

- python-dateutil=2.8.2=pyhd3eb1b0_0
- pyvis=0.3.1=pyhd8ed1ab_0 # added this line
- qt-main=5.15.2=h53bd1ea_10

Also: Did you take #42 into account for the new environment? It seems like you still pip install mlrose in install_packages.sh. For me it only works with the git repo pip install git+https://github.com/gkhayes/mlrose. I'm not sure about installing git repos through packages.yml, but installing with pip in the activated repository works just fine.

These packages were missing as they were not installed into my local environment by anaconda because it found them already installed in my global python version. With the last commit the packages should be updated. There are currently 3 installation possibilities: (1) packages.yml should contain all packages necessary to run the pipeline without graph-tool. (2) packages_wsbm.yml should contain all packages (including graph-tool). If this works, we will only keep the packages_wsbm.yml and remove the other one. The installation of mlrose will likely fail for (1) and (2) and should done independently as you proposed above using pip the repository directly. (3) As described in the readme, run source install_packages.sh.

Could you please test all 3 options?

[Garrafao]

With respect to the graph tool I suggest doing the following:

* create a new env with the graph tool `conda create --name wug -c conda-forge graph-tool`

* then use the .yml as follows `conda env update --file packages.yml`

If installation with packages_wsbm.yml doesn't work (see above), I will choose this as default installation recommendation.

[Garrafao]

Update: I updated both .yml files and the installation script with the last commit as pygraphviz was missing.