Fix bug in retro_branching when done on reset

[dmitrySorokin]

• When retro_branching::extract is called first time it creates a tree without scores. If env is done on reset it will remove all nodes and will rise a error when accessing removed root node. My solution is to set scores when creating the tree.
• rename column_features -> variable_features to support the latest ecole/scip version
• modify config for retro so it will run

[cwfparsonson]

Hi @dmitrySorokin

Many thanks for the PR!

I am reluctant to merge your changes in case they cause deviations from the original paper's experiments, which I believe should run.

Did the error with the retro branch in reset occur when you tried to run the paper's experiments, or when you tried to use retro branching in your own code? If the latter, I would recommend that you use the 'cleaner' version of retro branching in the notebook tutorial, which you should just be able to copy-paste (with the SearchTree class from the other tutorial) into your own project. These tutorials are in notebooks/tutorials/

[dmitrySorokin]

Hi @cwfparsonson

Thanks for the answer!
I will take a look on the notebooks.
The error appeared when I run new experiments with your code.

[cwfparsonson]

Also notice that in the notebook example, we call extract() on the retro branching reward object before the environment loop starts (ie just after we reset the environment) - I suppose this should really be a separate reset() method rather than extract(), but you should extract() before starting the env loop. Would this fix the error you were finding with extract()?