v0.7.0

IntervalMDP v0.7.0

Diff since v0.6.0

Changelog

• Unify underlying storage as FactoredRobustMarkovDecisionProcess; also supports factored models.
• Improve performance for IMDPs on GPU via bit-magic indexing in bitonic sorting.
• Add customizable Bellman operator algorithms.
• Add probabilistic labelling to ProductProcess.
• Improve docs, especially for specifications.
• Improve show of FactoredRobustMarkovDecisionProcess, problems, etc. - now prints key types, statistics, and configurations rather than the full transition matrices that clutter the REPL with little meaningfulness.

Breaking changes

• With the new FactoredRobustMarkovDecisionProcess model as a general underlying model, it requires a fixed number of actions per state. This will be relaxed in later versions. Convenience constructors for IntervalMarkovDecisionProcess and IntervalMarkovChain are included, but IntervalMarkovDecisionProcess currently requires a fixed number of actions.
• Rename from IntervalProbabilities to IntervalAmbiguitySets.

Merged pull requests:

• Probabilistic Labelling for ProductProcesses (#89) (@JeffersonYeh)
• Factored RMDPs as a base model (#90) (@Zinoex)
• fIMDP on cuda + other stuff (#94) (@Zinoex)

Closed issues:

• GPU for OrthogonalIntervalMarkovDecisionProcesses (#55)
• Factored IMDPs (+ mixtures) as a general model (#84)

v0.6.0

IntervalMDP v0.6.0

Diff since v0.5.0

Changelog

• Removed bellman operator, workspaces from public interface (breaking change)
• Changed interface to conform to CommonSolve interface (breaking change)

Merged pull requests:

• Bellman operator for ProductProcess (#87) (@Zinoex)
• Use CommonSolve standardized interface (#88) (@Zinoex)

v0.5.0

IntervalMDP v0.5.0

Diff since v0.4.6

• Update kernels for the Bellman operator bellman! to support rational arithmetic for exact value iteration.
• The bellman operator bellman! now takes a model rather than a probability object and a state pointer (breaking). The old interface is used underlying in _bellman_helper!, which is called by bellman!.

Merged pull requests:

• New Bellman interface (#85) (@Zinoex)

Closed issues:

• Rational number support (#14)

v0.4.6

IntervalMDP v0.4.6

Diff since v0.4.5

Merged pull requests:

• Wrong remainder? (#80) (@adrienbanse)
• Add initial ProductIMDP framework (#81) (@JeffersonYeh)
• Add expected exit time specification (#83) (@Zinoex)

v0.4.5

IntervalMDP v0.4.5

Diff since v0.4.4

v0.4.4

IntervalMDP v0.4.4

Diff since v0.4.3

v0.4.3

IntervalMDP v0.4.3

Diff since v0.4.2

v0.4.2

IntervalMDP v0.4.2

Diff since v0.4.1

Merged pull requests:

• Implicit sink state (#79) (@Zinoex)

Closed issues:

• Avoid specifying transition probabilities for sink states (#78)

v0.4.1

IntervalMDP v0.4.1

Diff since v0.4.0

Merged pull requests:

• Optimize PRISM data loading (#77) (@Zinoex)

v0.4.0

IntervalMDP v0.4.0

Diff since v0.3.0

Merged pull requests:

• Check device upload problem construction for better error messages (#42) (@Zinoex)
• Time-varying markov process (#43) (@Zinoex)
• Move the models into separate files (#44) (@Zinoex)
• CompatHelper: add new compat entry for Adapt in [weakdeps] at version 4, (keep existing compat) (#45) (@github-actions[bot])
• CompatHelper: add new compat entry for LLVM in [weakdeps] at version 7, (keep existing compat) (#46) (@github-actions[bot])
• CompatHelper: add new compat entry for GPUArrays in [weakdeps] at version 10, (keep existing compat) (#47) (@github-actions[bot])
• Introduce workspaces (#48) (@Zinoex)
• Massive refactor - primarily with the intention of incorporating ParallelProduct (#50) (@Zinoex)
• CompatHelper: bump compat for LLVM in [weakdeps] to 8, (keep existing compat) (#51) (@github-actions[bot])
• CompatHelper: bump compat for LLVM in [weakdeps] to 9, (keep existing compat) (#52) (@github-actions[bot])
• Remove ParallelProduct (#53) (@Zinoex)
• Orthogonal IMDPs (#54) (@Zinoex)
• More workspace types for OrthogonalIntervalProbabilities (#59) (@Zinoex)
• Simplify action vectors (#62) (@Zinoex)
• Refactor bellman! to prepare for mixture models (#63) (@Zinoex)
• Change interface for time varying models (#66) (@Zinoex)
• Add safety property (#67) (@Zinoex)
• CompatHelper: bump compat for GPUArrays in [weakdeps] to 11, (keep existing compat) (#68) (@github-actions[bot])
• Mixture model (#69) (@Zinoex)
• Use abs in convergence check (#72) (@bkraske)
• Fix strategy interface for CUDA (#75) (@Zinoex)
• Update docs (#76) (@Zinoex)

Closed issues:

• SparseOrdering to CuSparseOrdering adaptor (#12)
• Read bmdp-tool into an IntervalMarkovChain if num_actions == 1 (#30)
• Apply policy to IMDP to obtain IMC (#33)
• Test asserts (#35)
• Composition of IntervalMarkovProcesses (#38)
• Improve device error messages (#40)
• Test device check (#49)
• Simplify action vectors (#56)
• Sparse OrthogonalIntervalMarkovDecisionProcess (#57)
• Multi-threaded OrthogonalIntervalMarkovDecisionProcess (#58)
• Improve how strategies are applied (#64)
• Function to transfer data back to cpu (#74)