A solver for counterfactual inference over probabilistic logic programs.
WhatIf based on the aspmc library for probabilistic logic programming inference.
Its main functionality is the translation of counterfactual queries to marginal queries.
For usage on Linux you may install this software as a pip package via
pip install counterfactuals
Examples for command line usage are available below.
If you have any issues please contact us, or even better create an issue on GitHub.
For academic usage cite
- Kiesel, R., Rückschloß, K., & Weitkämper, F. (2023, July). "What if?" in Probabilistic Logic Programming. In Proceedings of the 39th International Conference on Logic Programming.
For developement clone via
git clone git@github.com:raki123/counterfactuals.git
We require Python >= 3.6.
All required modules are listed in requirements.txt and can be obtained by running
pip install -r requirements.txt
To use WhatIf as usual but have changes to the code available run
pip install -e .
in the root directory of this repository.
The basic usage is
WhatIf [-e .] [-ds .] [-dt .] [-k .] [-v .] [-h] [<INPUT-FILES>]
--knowlege -k COMPILER set the knowledge compiler to COMPILER:
* sharpsat-td : uses a compilation version of sharpsat-td (default)
* d4 : uses the (slightly modified) d4 compiler.
* c2d : uses the c2d compiler.
* miniC2D : uses the miniC2D compiler.
* pysdd : uses the PySDD compiler.
--evidence -e NAME,VALUE add evidence NAME:
* the evidence is not negated if VALUE is `True`.
* the evidence is negated if VALUE is `False`.
--intervene -i NAME,VALUE intervene on NAME:
* the intervention is not negated if VALUE is `True`.
* the intervention is negated if VALUE is `False`.
--query -q NAME query for the probability of NAME.
--decos -ds SOLVER set the solver that computes tree decompositions to SOLVER:
* flow-cutter : uses flow_cutter_pace17 (default)
--decot -dt SECONDS set the timeout for computing tree decompositions to SECONDS (default: 1)
--verbosity -v VERBOSITY set the logging level to VERBOSITY:
* debug : print everything
* info : print as usual
* result : only print results, warnings and errors
* warning : only print warnings and errors
* errors : only print errors
--help -h print this help and exit
When using the pip package replace python main.py by WhatIf to obtain the same result.
python main.py -q slippery -e sprinkler,True -i sprinkler,False -k sharpsat-td
0.5::u1.
0.7::u2.
0.1::u3.
0.6::u4.
szn_spr_sum :- u1.
sprinkler :- szn_spr_sum, u2.
rain :- szn_spr_sum, u3.
rain :- \+szn_spr_sum, u4.
wet :- rain.
wet :- sprinkler.
slippery :- wet.
Reads the sprinkler program from stdin and adds evidence sprinkler and intervention \+sprinkler.
The query is for slippery.
This results in the output
[WARNING] aspmc: Query for atom true was proven true during grounding.
[WARNING] aspmc: Including it has a negative impact on performance.
[INFO] aspmc: Tree Decomposition #bags: 18 unfolded treewidth: 3 #vertices: 20
[INFO] aspmc: Preprocessing disabled
[INFO] aspmc: Stats Compilation
[INFO] aspmc: ------------------------------------------------------------
[INFO] aspmc: Compilation time: 0.005887508392333984
[INFO] aspmc: Counting time: 0.0001952648162841797
[INFO] aspmc: ------------------------------------------------------------
[INFO] WhatIf: Results
[INFO] WhatIf: ------------------------------------------------------------
[RESULT] WhatIf: slippery: 0.09999999999999999
telling us that the result of the counterfactual query for slippery is 0.1.
The first two lines are a warning from aspmc that tell us that the atom true that we included to compute the probability of the evidence may lead to decreased performance. However, we need to include it as its probability is not 1.0 in general.
The following info lines tell us some stats about the program and the inference:
- it has a treewidth upper bound of 3
- aspmc's preprocessing is disabled
- knowledge compilation took ~0.006 seconds
- counting over the resulting circuit took ~0.0002 seconds