Pseudo-branching #190
Pseudo-branching #190
Conversation
…the similarl named examples already existing
…g. Current version runs on the two examples however it only uses MOST_INFEASIBLE, i.e. costs have not stabilized.
…et. I.e. this is more like a non sorted collections of potential sections
… from approx planted example, also added a \mu for a potential branching score function
…rt. Current version runs but on the HiGHS_example with diffw = Random.rand(Bool, n) * 0.6 .+ 0.3 it does not change strategy before solving the problem.
…pseudocost branching
…seudocost branch.
…alling optimize. Before having them in get_branching_variable caused them to be reset on every call which made a strategy switch impossible
…he instance generated)
Modify tree
|
This version seems to work. It has been tested on the sparse_regression example where occasionaly decisions are made based on pseudocosts. I have yet to observe an instance where after some time all decisions are made on pseudocost knowledge, i.e. some variables don´t meet the requirement even after other variables have been branched on tens of times. Requirement for testing was that a variable has been branched up and down at least once. On the Approximate planted point example current code does not work yet when CubeSimpleBLMO is passed. Test Summary: | Pass Error Total Time |
…w updates to be made scaled to unit cost. Previously this was done incorrectly
…ates are scaled. the previous version scaled updates incorrectly resulting in only Inf value updates.
…it is even created
hierarchical pseudobranching strategy
|
Gradient based branching not working yet. I´m not sure why it doesn´t find the method. |
…anching related ones to a new script called branchings_strategies.jl
|
It would also be nice to have some examples and tests to check that the new branching strategies work as intended and showcase their usefulness. |
Only for hierarchy so far to test if it will still work.
…into the branching strategy which now is a mutable structure
No description provided.