Maxime Gasse

I like the idea, makes the code more explicit as to when reward / observation functions are called. That could apply as well to observation functions right ? Here are...

Hi @only-changer , We have students working on those topics here, but nothing ripe enough to be integrated into Ecole. Regarding cut selection, please note that SCIP 8 offers a...

Good, I had not realized that ! Typically I want to be able to catch an exception triggered from SCIP, and add more information / context to it. Say, I...

In my last experiments it seems pybind does not process nested exceptions the way I want, as it only prints the lastly thrown exception... Maybe not the way to go...

Wow, that looks like big progress. Well done @AntoinePrv . Let's discuss what to do with that in the next meeting.

Hmmm looks like the tests do fail... ``` def test_extract(observation_function, model): """Obtain observation.""" observation_function.before_reset(model) pytest.helpers.advance_to_stage(model, ecole.scip.Stage.Solving) > observation_function.extract(model, False) E ecole.core.scip.Exception: Constraint 0 cannot be expressed as a single linear...

SCIP's default policy is not really compatible with fair node counting, as it has many side-effects besides taking branching decisions. Tracking all those side-effects (mostly variable tightenings) to compute a...

We can further improve the code, and get rid of the vector altogether by accumulating the integral on-the-fly without storing all intermediate bound values. Let's keep this issue open until...

I would answer yes to the three questions: - I would have just one branching environment. Existing code for single-variable branching will not change, so MVB should be backward-compatible with...

I'm really not sure, no. The code looks right to me, however when using it it would often bring SCIP to inconsistent states, such as unfeasibility or optimality with incorrect...