v0.8.1

Minor bugfix, missing c++ import 7472a00

v0.8.0

New features

• Make libecole a installable CMake library (no Python). #240
• Refactor Ecole packaging to allow the creation of Ecole Python extensions. #282
• Add method to reactivate SCIP logger. #283
• Added possibility use SCIP default in Branching environment by passing ecole.Default #294 #303

Bug fixes

• Normalized edge coefficients in node bipartite observation. #270
• Fix difficulty of instance generators #301

Breaking changes

• Change StrongBranching pseudo_candidate parameter default value. #239
• Observations functions now use uniformly variable indices (instead of LP). In most cases, this will not impact user code. #210
• Remove deprecated functions/emum NodeBipartiteObs.column_features, NodeBipartiteObs.ColumnFeatures
• Rename RandomEngine to RandomGenerator
• Renamed exceptions to ScipError and MarkovError #302
• Move to SCIP8 on Conda-forge #303

v0.7.3

Bug fixes

• Add constructor option to extract only LP branching candidates in Kalild2016 (default behavior is changed to match that of the Branching environment).

v0.7.2

Bug fixes

• Reward / observation extraction is now made in the proper order in both the C++ and Python Environment class
• Observations are not extracted twice any more in the Python Environment class (bug introduced in v0.7.0)

v0.7.1

Bug fixes

• Fix discrepancy between Ecole build-time version for libfmt (fmt7) and the conda package dependency version for libfmt (fmt8)

v0.7.0

New features

• Add a FileGenerator to iterate over local problem instances #174
• Add Hutter2011 observation function #192
• Add PrimalIntegral, DualIntegral, PrimalDualIntegral reward functions #144
• Add BranchingSum multi-variable branching environment #181
• Add PrimalSearch environment #183
• Terminal states (done==True) always return None as an observation #199
• Rewards are now extracted before observations, to account for the observation extraction cost between two rewards #212

Bug fixes

• Fixed a bug in IndependentSetGenerator where the greedy clique strengthening was not as greedy as it should.
• Documentation typos

v0.6.2

Bug fixes

• Fix invalid NodeBipartite when constraint had both lhs and rhs #189

v0.6.1

New features

• Add PyPI source package #185 #37

v0.6.0

New features

• Add MilpBipartite observation function for pre-solving (#147)
• Add Pickle support to all observation functions (#106 #152)
• Features description for Khalil2016 (#157)

Breaking changes

• NodeBipartite features have been reordered
• Changed combinatorial auctions, maximum independent set and capacitated facility location instance generators so that they produce instances of difficulty similar to those of Gasse et al. (2019) out-of-the-box (#141, #142 #150)
• Khalil2016 now returns a struct wrapping the Numpy array (#157)

v0.5.1

Bug fixes

• Fix performance bottleneck in Khalil2016
• Improve Python detection in source (venv) builds #133