See installation instructions here.
Although we share the instances used for our experiments, you can generate your own instances.
We provide a script in the ./scripts/ folder to generate instances for the ecole benchmark.
# Generate MILP instances
sh generate_instances.py
For a given TYPE in {SetCovering, CombinatorialAuction, MaximumIndependentSet, MultipleKnapsack},
and a given AGENT in {dqn_bbmdp, dqn_retro, dqn_tMDP, pg_tMDP, il}, run scripts in the ./scripts/
folder to reproduce experiments on the Ecole benchmark.
# Evaluation
sh run_benchmark.sh $TYPE $AGENT
# Training (only for reinforcement learning DQN agents)
sh run_dqn_trainer.sh $TYPE $AGENT
Both evaluation and training performance can be monitored from the folder located in ./experiments/logs/$TYPE/.