Page Contributors: Adam Green,
Planner Quality: 31.2 out of 100
Year Published: 2014
Paper: IPC 2014 Booklet [ Malitsky, Y. Wang, D. Karpas, E. ]
Preceded By: Fast Downward
The AllPACA planner is a portfolio planner, which automatically chooses which of several planners to run for the planning task that it is given. AllPACA is based on machine learning techniques, which attempt to choose the planner that will result in the fastest solution time, based on the features of the planning task. In the sequential optimal track, AllPACA was pre-trained on all planning tasks the sequential optimal track in all previous editions of the International Planning Competition.
For the learning track, AllPACA can also learn to predict planner performance on tasks from each domain during the training phase, and can additionally exploit domainspecific features.
AllPACA has not been tested with eviscerator as we couldn’t get the source code to compile. Fast Downward forms the base planner of AllPACA, please refer to the Fast Downward [Link Needed] page for an idea of the supported features in this planner
Source code for AllPACA was zipped with all IPC 2014 submissions here