Code and Environment for Causal Induction From Visual Observations for Goal-Directed Tasks.
Consists of the light switch environment for studying visual causal induction, where N switches control N lights, under various causal structures. Includes common cause, common effect, and causal chain relationships. Environment code resides under env/light_env.py
.
The different induction models used are located under F_models.py
, incuding our proposed iterative attention network, as well as baselines which do not use attention or use temporal convolutions.
Step 1: Generate Data
python3 collectdata.py --horizon 7 --num 7 --fixed-goal 0 --structure one_to_one --seen 10 --images 1 --data-dir output/
Step 2: Train Induction Model
python3 trainF.py --horizon 7 --num 7 --fixed-goal 0 --structure one_to_one --type iter --images 1 --seen 10 --data-dir output/
Step 3: Eval Induction Model
python3 evalF.py --horizon 7 --num 7 --fixed-goal 0 --structure one_to_one --method trajFi --images 1 --seen 10 --data-dir output/
Step 4: Train Policy via Imitation
python3 learn_planner.py --horizon 7 --num 7 --fixed-goal 0 --structure one_to_one --method trajFi --seen 10 --images 1 --data-dir output/