gautamsgev/ELectrical_Metaopenenv

┌─────────────┐    action      ┌───────────────────┐   SPICE netlist  ┌──────────────┐
│  RL Policy  │ ──────────→   │  CircuitSynthEnv  │ ──────────────→  │   ngspice    │
│ (any algo)  │               │  (OpenEnv API)    │                  │  (simulator) │
└─────────────┘ ←──────────── └───────────────────┘ ←──────────────  └──────────────┘
                 obs + reward        circuit graph         waveform data

git clone https://github.com/your-org/circuitsynth-squarewave
cd circuitsynth-squarewave
pip install -e .

sudo apt-get install ngspice        # Ubuntu / Debian
brew install ngspice                # macOS (Homebrew)

from circuitsynth import CircuitSynthEnv

# Create environment (mock mode — no ngspice required)
env = CircuitSynthEnv(task_id="squarewave-easy", seed=42, mock_sim=True)

# OpenEnv API
obs, info = env.reset()

for step in range(30):
    action = env.action_space.sample()          # random policy
    obs, reward, terminated, truncated, info = env.step(action)
    print(f"Step {step}: reward={reward:.4f}  components={info['n_components']}")
    if terminated or truncated:
        break

# Full MDP state (for debugging / logging)
state = env.state()
print(state["reward_decomposition"])

# Use dict-style actions for scripting / debugging
obs, info = env.reset()

env.step_dict({"action_type": "ADD_COMPONENT", "component_type": "VSOURCE",
               "value_idx": 5, "node_a": "VCC", "node_b": "GND"})
env.step_dict({"action_type": "ADD_COMPONENT", "component_type": "RESISTOR",
               "value_idx": 5, "node_a": "VCC", "node_b": "N1"})
env.step_dict({"action_type": "ADD_COMPONENT", "component_type": "NPN_BJT",
               "value_idx": 0, "node_a": "N1", "node_b": "N3", "node_c": "GND"})
# ... build full circuit ...

obs, reward, terminated, truncated, info = env.step_dict({"action_type": "FINALIZE"})
print(f"Final reward: {reward:.4f}")
print(f"Frequency: {info['waveform_metrics']['frequency']:.1f} Hz")

VCC ─┬── R1(1kΩ) ──┬── C1(15nF) ──┬── R4(47kΩ) ──┐
     │             N1              N4              │
     │              ├── Q1(C)     Q2(B) ──┐       │
     └── R2(1kΩ) ──┬── C2(15nF) ──┬── R3(47kΩ)  │
                   N2              N3              │
                    └── Q2(C)     Q1(B) ──┘       │
Q1.E, Q2.E → GND                                  │
Output at N1 (or N2)                               │

python scripts/evaluate.py --reference --task squarewave-easy --mock

from stable_baselines3 import PPO
from circuitsynth import CircuitSynthEnv

env = CircuitSynthEnv(task_id="squarewave-easy", mock_sim=True)

model = PPO(
    "MlpPolicy", env,
    learning_rate=3e-4,
    n_steps=2048,
    batch_size=64,
    verbose=1,
)
model.learn(total_timesteps=500_000)
model.save("circuitsynth_ppo")

# Evaluate
obs, _ = env.reset()
for _ in range(30):
    action, _ = model.predict(obs)
    obs, r, done, trunc, info = env.step(action)
    if done or trunc:
        break
print(f"Reward: {r:.4f}  Freq: {info['waveform_metrics']['frequency']:.1f} Hz")

for task_id in ["squarewave-easy", "squarewave-medium", "squarewave-hard"]:
    env = CircuitSynthEnv(task_id=task_id, mock_sim=False)
    model = PPO.load("circuitsynth_ppo")
    model.set_env(env)
    model.learn(total_timesteps=200_000)
    model.save(f"circuitsynth_ppo_{task_id}")

# Random baseline (no ngspice needed)
python scripts/baseline_inference.py --task squarewave-easy --episodes 20 --mock

# Reference circuit reward validation
python scripts/evaluate.py --reference --task squarewave-easy --mock

# Evaluate a trained SB3 model
python scripts/evaluate.py --model circuitsynth_ppo.zip --task squarewave-hard

# Run tests
pytest tests/ -v

circuitsynth-squarewave/
├── README.md
├── Dockerfile              # HuggingFace Spaces compatible
├── openenv.yaml            # OpenEnv specification
├── requirements.txt
├── setup.py
│
├── circuitsynth/
│   ├── __init__.py         # Package API
│   ├── env.py              # CircuitSynthEnv (main class)
│   ├── tasks.py            # Task registry (easy / medium / hard)
│   ├── action_space.py     # Typed actions + MultiDiscrete encoding
│   ├── observation.py      # Observation builder (flat + graph)
│   ├── reward.py           # Decomposed reward function
│   ├── netlist.py          # Netlist graph + SPICE serializer + validation
│   ├── simulator.py        # ngspice subprocess wrapper + mock fallback
│   ├── waveform.py         # Waveform analysis (FFT, DTW, metrics)
│   ├── components.py       # Component library + SPICE models
│   └── utils.py            # Helpers: seeding, SI formatting, convergence detection
│
├── scripts/
│   ├── baseline_inference.py   # Reproducible random policy rollout
│   └── evaluate.py             # Offline evaluation + reference circuit
│
└── tests/
    ├── test_env.py
    ├── test_reward.py
    ├── test_simulator.py
    └── test_waveform.py