Code release for ECAI 2024 paper "Decision-Focused Learning to Predict Action Costs for Planning"

Fast Downward Installation Intructions

Inside the repository

git clone https://github.com/aibasel/downward.git
cd downward
./build.py

Virtual Environment Creation

python3 -m venv env_dflforplanning
source env_dflforplanning/bin/activate
pip install -r requirements.txt

Data Generation

Create directoy mkdir data Then generate the datsets by runnning gen_data.sh

• Refer to DescriptionofInstances.txt to find which .sas file to use for running each model mentioned in the paper.

Running The Experiments

Run Exp_run.sh to run different configurations with A* with LM-Cut.

• To run with WA* with LM-Cut add --method 'wastar'
• To run with GBFS with hFF add --method 'ffh'
• To Run with caching add --caching --psolve 0.2 (for p=20%)

Run ShortestPathExp_run.sh to run the shortestpath experiments.

Tabulating Results

For tabulating results you may use plot_roversresults.py, plot_roversresults.py, plot_shortestpath.py.