Tim Quatmann
|
ab95e7d08b
|
BeliefManager: organized stored beliefs in buckets (beliefs with the same observation belong in the same bucket)
|
6 years ago |
Sebastian Junges
|
e9e9b15cb1
|
store/load winning region to file
|
6 years ago |
|
Sebastian Junges
|
b2e7c5d5ed
|
various changes to allow restarting and more finegrained selection of switch-and-finish-with-policy
|
6 years ago |
|
Sebastian Junges
|
43bb70e93d
|
bugfix where the wrong successor variables where selected
|
6 years ago |
|
Sebastian Junges
|
5783719c05
|
add a validator to the winning region search
|
6 years ago |
|
Tim Quatmann
|
6f476ef079
|
belief exploration: Improved fixpoint detection for over-approx
|
6 years ago |
|
Tim Quatmann
|
ddec9ce740
|
ApproximatePomdpModelchecker: Fixed output a little.
|
6 years ago |
Matthias Volk
|
d3c8093e0f
|
Removed unnecessary semicolons
|
6 years ago |
|
Sebastian Junges
|
c0ac9814e1
|
allow for graph-analysis and sat-based analysis interleaving, and restarting sat-based solver when advantageous
|
6 years ago |
|
Sebastian Junges
|
ea73e246a7
|
A new qualitative reachability analysis for POMDPs/prob1max, based on graphs (sound but incomplete).
|
6 years ago |
|
TimQu
|
7504f6f315
|
Improved statistics output for refinements, added detection of fixpoints
|
6 years ago |
|
Tim Quatmann
|
5a76f7355d
|
Fixed an issue with refinement of under-approximation
|
6 years ago |
|
Tim Quatmann
|
8b4595042e
|
Only do iteration output if the result bound improved. Handle integer overflows for the observation resolution.
|
6 years ago |
|
Sebastian Junges
|
e4a4214fc4
|
towards more helpful output when POMDP is not canonical
|
6 years ago |
|
Sebastian Junges
|
0e97647d9e
|
canonic POMDPs are now annotated with the state valuations and the choice labellings
|
6 years ago |
|
Sebastian Junges
|
5277139640
|
first version of a generator for belief support MDPs
|
6 years ago |
|
Sebastian Junges
|
94c87a9926
|
first version of query interface that is more accessible than the winning region itself
|
6 years ago |
|
Sebastian Junges
|
86ece9d324
|
cleaning code
|
6 years ago |
|
Tim Quatmann
|
43220759f4
|
Implemented a time limit for exploration.
|
6 years ago |
|
Tim Quatmann
|
e81b8f1622
|
BeliefManager: fixed a few assertion conditions
|
6 years ago |
|
Tim Quatmann
|
1763f0c582
|
Making sure that we only store the best bounds found so far. Also added some output for the resulting values in each iteration.
|
6 years ago |
|
Tim Quatmann
|
fa624d2a20
|
Introduced new settings for controlling the refinement strategy and whether to produce only upper and/or lower bounds
|
6 years ago |
|
Tim Quatmann
|
2d94e77f2a
|
Only display the bound that was requested.
|
6 years ago |
|
Tim Quatmann
|
37490a8eca
|
Started to integrate new refinement options.
|
6 years ago |
|
Tim Quatmann
|
75d792e987
|
Implemented refinement heuristic.
|
6 years ago |
|
Tim Quatmann
|
45832d3de3
|
BeliefMdpExplorer: Implemented extraction of optimal scheduler choices and reachable states under these choices
|
6 years ago |
|
Tim Quatmann
|
61215e4b24
|
Over-Approximation: Taking current values as new lower/upper bounds for next refinement step.
|
6 years ago |
|
Tim Quatmann
|
4ea452854f
|
Fixes for scoring observations
|
6 years ago |
Sebastian Junges
|
5f2a598f48
|
remove unsound 1-state computation
|
6 years ago |
|
Sebastian Junges
|
6608f9f00d
|
Fixed implementation from CCD16
|
6 years ago |
|
Sebastian Junges
|
7ba3b6b8d6
|
Canonic POMDP in -> Canonic POMDP out
|
6 years ago |
|
Sebastian Junges
|
e22cbdb91b
|
support for computing the winning region or from initial state, some documentation
|
6 years ago |
|
Tim Quatmann
|
26764137f5
|
Fix for --unfold-belief-mdp setting
|
6 years ago |
|
Tim Quatmann
|
3c5df045c1
|
Added a few assertions
|
6 years ago |
|
Tim Quatmann
|
03889958da
|
Added a switch to control the size of the under-approximation via command line.
|
6 years ago |
|
Tim Quatmann
|
26a0544e4b
|
BeiliefManager: Use flat_maps for beliefs and hash_maps for belief storage.
|
6 years ago |
|
Tim Quatmann
|
fcee1d05fa
|
Fixed an issue with dropping unexplored states.
|
6 years ago |
|
Tim Quatmann
|
2f020ce686
|
BeliefManager: Making Freudenthal happy (and fast)
|
6 years ago |
|
Tim Quatmann
|
937659f356
|
First improvement step for Freudenthal triangulation
|
6 years ago |
|
Tim Quatmann
|
eca4dab6c0
|
Beliefmanager: expanding a belief now returns a vector instead of a map
|
6 years ago |
|
Tim Quatmann
|
26864067cf
|
BeliefManager: Made several methods private to hide the actual BeliefType.
|
6 years ago |
|
Tim Quatmann
|
5cd4281133
|
Further output improvements.
|
6 years ago |
|
Tim Quatmann
|
34d6ac9fe1
|
Fixed computing a state limit for the under-approximation.
|
6 years ago |
|
Tim Quatmann
|
961baa4386
|
BeliefMdpExplorer: Various bugfixes for exploration restarts. Unexplored (= unreachable) states are now dropped before building the MDP since we do not get a valid MDP otherwise.
|
6 years ago |
|
Tim Quatmann
|
c2837bb749
|
ApproximatePOMDPModelchecker: Improved output a bit.
|
6 years ago |
|
Tim Quatmann
|
c3847d05af
|
Scaling the rating of an observation with the current resolution.
|
6 years ago |
|
Tim Quatmann
|
c2ddea1480
|
First (re-) implementation of refinement. (probably needs some testing/debugging)
|
6 years ago |
Alexander Bork
|
62c905fc58
|
Added basis for rewards in dropUnreachableStates()
|
6 years ago |
|
Alexander Bork
|
3041b881d4
|
Beginning of dropUnreachableStates()
|
6 years ago |
|
Tim Quatmann
|
79641ef131
|
Started to make the BeliefMdpExplorer more flexible, allowing to restart the exploration
|
6 years ago |
