Makefile

.PHONY: help env weights train-safety train-task cpo collect figures reproduce clean
.DEFAULT_GOAL := help

# ------------------------------------------------------------------ knobs ----
ENV     ?= SafetyCarCircle2-v0      # SafetyCarCircle2-v0 | SafetyCarGoal1-v0
CFG     ?= configs/task/circle_rollout.yaml
SEED    ?= 0
DATA    ?= data/results
FIGDIR  ?= figures

help:
	@echo "PORL reproduction targets:"
	@echo "  make env           - print the one-time install steps"
	@echo "  make weights       - download published safety-policy weights"
	@echo "  make train-safety CFG=configs/safety_policy/circle.yaml"
	@echo "  make train-task    CFG=configs/task/circle_rollout.yaml SEED=0"
	@echo "  make cpo           ENV=SafetyCarCircle2-v0 SEED=0"
	@echo "  make collect       - rebuild data/results/*.csv from local runs (retrain path)"
	@echo "  make figures       - regenerate figures/ from committed data/results/ (no GPU)"
	@echo "  make reproduce     - full retrain -> collect -> figures (slow, needs GPU)"

env:
	@echo "conda create -n porl python=3.10 -y && conda activate porl"
	@echo "pip install -e external/safety-stable-baselines"
	@echo "pip install -e external/safety-stable-baselines/integrations/safety-gymnasium"
	@echo "pip install -e external/Safe-Policy-Optimization"
	@echo "pip install -e ."

weights:
	bash scripts/fetch_weights.sh

train-safety:
	python -m porl.train_safety_policy --config $(CFG) --seed $(SEED)

train-task:
	python -m porl.train --config $(CFG) --seed $(SEED)

cpo:
	python -m porl.run_cpo --env $(ENV) --seed $(SEED)

collect:
	python -m analysis.collect_results --out-dir $(DATA)

figures:
	python -m analysis.plot_figures --data-dir $(DATA) --out-dir $(FIGDIR)

# Full end-to-end reproduction. Trains the 2 safety policies, then 5 seeds x 2 envs x
# {naive, value, rollout} task policies + CPO, collects, and plots. Expect many GPU-hours.
reproduce:
	python -m porl.train_safety_policy --config configs/safety_policy/circle.yaml
	python -m porl.train_safety_policy --config configs/safety_policy/goal.yaml
	@for env in circle goal; do \
	  for method in naive value rollout; do \
	    for s in 0 1 2 3 4; do \
	      python -m porl.train --config configs/task/$$env\_$$method.yaml --seed $$s ; \
	    done ; \
	  done ; \
	done
	@for envid in SafetyCarCircle2-v0 SafetyCarGoal1-v0; do \
	  for s in 0 1 2 3 4; do python -m porl.run_cpo --env $$envid --seed $$s ; done ; \
	done
	$(MAKE) collect
	$(MAKE) figures

clean:
	rm -rf $(FIGDIR)/*.pdf $(FIGDIR)/*.svg $(FIGDIR)/*.png