Spaces:
Sleeping
Sleeping
File size: 14,710 Bytes
9b50e1b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 | #!/usr/bin/env python3
"""
scripts/evaluate_cli.py β Terminal Human Agent Evaluator
=========================================================
Play through the irrigation environment in the terminal, OR submit a
pre-made action sequence, and receive programmatic + LLM evaluation.
Usage:
# Interactive (default env):
python scripts/evaluate_cli.py --mode interactive
# Interactive with drought preset:
python scripts/evaluate_cli.py --mode interactive --preset drought
# Batch mode β comma-separated actions:
python scripts/evaluate_cli.py --mode batch --actions "0,2,2,0,1,3,0,2,0,0"
# Batch with heavy drought + extra sensor noise:
python scripts/evaluate_cli.py --mode batch --actions "2,3,2,3,2" \
--preset drought --rain-mult 0.1 --sensor-noise 0.08
# Load a saved JSON config:
python scripts/evaluate_cli.py --mode batch --actions "0,2,2" --config-file my_env.json
# Dump active config as JSON (for saving / editing):
python scripts/evaluate_cli.py --preset hard --dump-config
# List all presets:
python scripts/evaluate_cli.py --list-presets
# Enable Claude AI coaching:
export ANTHROPIC_API_KEY=sk-ant-...
python scripts/evaluate_cli.py --mode interactive
"""
from __future__ import annotations
import argparse, json, os, sys, time
import numpy as np
sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
from envs.env_config import EnvConfig
from envs.configurable_env import make_env
from evaluator.programmatic_grader import ProgrammaticGrader
from evaluator.llm_grader import LLMGrader
ACTION_LABELS = {0:"No Water", 1:"Low (1kL)", 2:"Medium (3kL)", 3:"High (5kL)", 4:"Drain"}
STAGE_NAMES = {0:"Seedling", 1:"Vegetative", 2:"Reproductive", 3:"Maturity"}
WATER_VOLS = [0, 1000, 3000, 5000, 0]
BANNER = """
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β πΎ Precision Irrigation RL Β· Terminal Evaluator β
β OpenEnv Hackathon Β· Meta Γ Hugging Face β
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
"""
# ββ Display helpers ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _bar(v, width=20, lo=0.0, hi=1.0):
b = "β"*int(v*width) + "β"*(width-int(v*width))
return f"[{b}] {'β' if lo<=v<=hi else 'β'}"
def _show_state(obs, day, last_r, total_r, cfg):
m = float(obs["soil_moisture_obs"][0])
cg = float(obs["crop_growth"][0])
cs = int(round(float(obs["crop_stage"][0])))
wt = float(obs["water_tank"][0])
fc = obs["weather_forecast"]
status = "OPTIMAL β
" if cfg.opt_lo<=m<=cfg.opt_hi else ("DRY π΄" if m<cfg.opt_lo else "WET π§")
print(f"\n{'β'*62}")
print(f" Day {day:>3}/{cfg.max_steps} Last reward: {last_r:+.2f} Total: {total_r:+.2f}")
print(f"{'β'*62}")
print(f" Soil Moisture : {m:.3f} {_bar(m,lo=cfg.opt_lo,hi=cfg.opt_hi)} {status}")
print(f" Crop Growth : {cg:.3f} {_bar(cg)}")
print(f" Growth Stage : {STAGE_NAMES.get(cs, str(cs))}")
print(f" Water Tank : {wt:.1%} (capacity={cfg.tank_capacity/1000:.0f}kL)")
rain_fc = " | ".join(f"{fc[i,0]:.1f}mm/{fc[i,1]:.0f}Β°C" for i in range(3))
print(f" 3-Day Forecast: {rain_fc}")
print(f"{'β'*62}")
def _menu():
print("\n Actions:")
for k,v in ACTION_LABELS.items():
print(f" [{k}] {v}")
print(" [h] Get PPO hint")
print(" [q] Quit early")
def _human_action(obs):
while True:
raw = input(" Choose > ").strip().lower()
if raw == "q": return None
if raw == "h": return "hint"
try:
a = int(raw)
if 0 <= a <= 4: return a
except ValueError:
pass
print(" β Enter 0-4, h, or q.")
# ββ Episode runners ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def run_interactive(cfg: EnvConfig, seed=42):
print(BANNER)
print(f" Config : {cfg.summary()}")
print(f" Optimal moisture band: [{cfg.opt_lo}, {cfg.opt_hi}]")
input("\n Press Enter to start...\n")
from rl.obs_normalizer import manual_normalize
from rl.ppo import ActorCritic, softmax
import pickle, pathlib
ppo = None
pkl = pathlib.Path(__file__).parent.parent / "models" / "ppo_irrigation.pkl"
if pkl.exists():
try:
# Preferred path for current PPO checkpoints (state dict).
ppo = ActorCritic.load(str(pkl))
except Exception:
# Backward compatibility for older pickled object checkpoints.
with open(pkl, "rb") as f:
ppo = pickle.load(f)
def _ppo_action(obs):
if ppo is None:
return None
x = manual_normalize(obs)
# Current API: ActorCritic.forward returns (logits, value).
if hasattr(ppo, "forward"):
logits, _ = ppo.forward(x[np.newaxis])
return int(np.argmax(softmax(logits[0])))
# Legacy compatibility path (older custom actor wrappers).
if hasattr(ppo, "actor"):
return int(np.argmax(softmax(ppo.actor(x))))
return None
env = make_env(cfg, seed=seed)
obs, _ = env.reset(seed=seed)
history, total_r, last_r = [], 0.0, 0.0
for step in range(1, cfg.max_steps+1):
_show_state(obs, step, last_r, total_r, cfg)
_menu()
action = _human_action(obs)
if action is None:
print("\nβ οΈ Quit early."); break
if action == "hint":
try:
ha = _ppo_action(obs)
except Exception:
ha = None
print(f" π€ PPO β {ACTION_LABELS.get(ha,'N/A')}" if ha is not None else " π€ PPO model not loaded.")
action = _human_action(obs)
if action is None: break
obs, reward, terminated, truncated, info = env.step(action)
last_r = reward; total_r += reward
history.append({
"day":step,"action":action,
"moisture":float(obs["soil_moisture_obs"][0]),
"crop_growth":float(obs["crop_growth"][0]),
"crop_stage":int(round(float(obs["crop_stage"][0]))),
"rainfall_mm":info.get("rain_mm",0.0),
"reward":float(reward),
"tank_level":float(obs["water_tank"][0]),
"water_used_L":max(0,WATER_VOLS[action]),
})
print(f" β {ACTION_LABELS[action]:14s} β Reward: {reward:+.2f} Total: {total_r:+.2f}")
if info.get("flash_flood"): print(" β οΈ Flash flood!")
if info.get("power_cut"): print(" β‘ Power cut β pump unavailable today!")
if terminated or truncated:
print(f"\nβ
Episode complete at Day {step}"); break
env.close()
return history
def run_batch(actions, cfg: EnvConfig, seed=42):
env = make_env(cfg, seed=seed)
obs, _ = env.reset(seed=seed)
history, total_r = [], 0.0
print(f"\n Running {len(actions)} actions (seed={seed})")
print(f" Config : {cfg.summary()}\n")
for step, action in enumerate(actions, 1):
action = max(0, min(4, action))
obs, reward, terminated, truncated, info = env.step(action)
total_r += reward
history.append({
"day":step,"action":action,
"moisture":float(obs["soil_moisture_obs"][0]),
"crop_growth":float(obs["crop_growth"][0]),
"crop_stage":int(round(float(obs["crop_stage"][0]))),
"rainfall_mm":info.get("rain_mm",0.0),
"reward":float(reward),
"tank_level":float(obs["water_tank"][0]),
"water_used_L":max(0,WATER_VOLS[action]),
})
if step % 15 == 0 or step <= 3:
m = float(obs["soil_moisture_obs"][0])
print(f" Day {step:>3}: action={action}({ACTION_LABELS[action]:<14}) "
f"moisture={m:.2f} reward={reward:+.2f} cumulative={total_r:+.2f}")
if terminated or truncated: break
env.close()
print(f"\n Completed {len(history)} steps | Total reward: {total_r:.2f}")
return history
# ββ Evaluation βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def evaluate_and_print(history, cfg: EnvConfig, api_key=None):
print(f"\n{'β'*62}")
print(f" EVALUATING YOUR PERFORMANCE ({len(history)} steps)")
print(f" Config: {cfg.summary()}")
print(f"{'β'*62}")
grader = ProgrammaticGrader(history)
prog = grader.evaluate()
print("\n" + ProgrammaticGrader.format_scores_text(prog))
key = api_key or os.environ.get("ANTHROPIC_API_KEY","")
llm = LLMGrader(api_key=key)
if llm.is_available:
print("\nπ€ Consulting Claude AI Coach...")
t0 = time.time()
result = llm.evaluate(history, prog)
print(f" (response in {time.time()-t0:.1f}s)\n")
else:
print("\nβΉοΈ No API key β using rule-based coaching (set ANTHROPIC_API_KEY for Claude analysis)")
result = llm.evaluate(history, prog)
print(LLMGrader.format_evaluation_text(result))
out_path = os.path.join(os.path.dirname(__file__),"..","outputs","human_eval.json")
os.makedirs(os.path.dirname(out_path), exist_ok=True)
with open(out_path,"w") as f:
json.dump({
"history_length":len(history),
"config":cfg.to_dict(),
"programmatic":prog,
"llm_evaluation":result,
}, f, indent=2)
print(f"\nπ Full results β {os.path.abspath(out_path)}")
# ββ Main βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def main():
parser = argparse.ArgumentParser(
description="Precision Irrigation RL β Terminal Evaluator",
formatter_class=argparse.RawTextHelpFormatter,
epilog="Available presets: " + ", ".join(EnvConfig.preset_names()),
)
parser.add_argument("--mode", choices=["interactive","batch"], default="interactive")
parser.add_argument("--actions", type=str, default=None)
parser.add_argument("--actions-file", type=str, default=None)
parser.add_argument("--seed", type=int, default=42)
parser.add_argument("--api-key", type=str, default=None)
# Config
parser.add_argument("--preset", type=str, default="normal",
help="Env preset (normal|drought|flood|power_outage|water_scarce|easy|hard|high_yield_challenge)")
parser.add_argument("--config-json", type=str, default=None,
help='JSON overrides, e.g. \'{"rain_multiplier":0.3,"et_coeff":0.007}\'')
parser.add_argument("--config-file", type=str, default=None,
help="Path to a saved JSON config file")
parser.add_argument("--dump-config", action="store_true",
help="Print active config JSON and exit")
parser.add_argument("--list-presets", action="store_true",
help="List all available presets and exit")
# Quick individual overrides
parser.add_argument("--rain-mult", type=float, default=None, help="Rain multiplier (0=drought, 2=flood)")
parser.add_argument("--et-coeff", type=float, default=None, help="Evapotranspiration coefficient")
parser.add_argument("--max-steps", type=int, default=None, help="Episode length in days")
parser.add_argument("--tank", type=float, default=None, help="Tank capacity (litres)")
parser.add_argument("--power-cut", type=float, default=None, help="Power cut probability 0β1")
parser.add_argument("--sensor-noise", type=float, default=None, help="Sensor noise std dev")
args = parser.parse_args()
if args.list_presets:
print("\nAvailable presets:\n")
for name in EnvConfig.preset_names():
cfg = EnvConfig.from_preset(name)
print(f" {name:<25} {cfg.summary()}")
return
# ββ Build config ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
if args.config_file:
cfg = EnvConfig.from_json(args.config_file)
else:
cfg = EnvConfig.from_preset(args.preset)
if args.config_json:
overrides = json.loads(args.config_json)
d = cfg.to_dict(); d.update(overrides)
cfg = EnvConfig.from_dict(d)
# Individual quick overrides
quick = {
"rain_multiplier": args.rain_mult,
"et_coeff": args.et_coeff,
"max_steps": args.max_steps,
"tank_capacity": args.tank,
"power_cut_prob": args.power_cut,
"sensor_noise_std": args.sensor_noise,
}
d = cfg.to_dict()
for k,v in quick.items():
if v is not None: d[k] = v
cfg = EnvConfig.from_dict(d)
if any(v is not None for v in quick.values()):
cfg.preset = "custom"
errs = cfg.validate()
if errs:
print("β Config validation errors:"); [print(f" {e}") for e in errs]; sys.exit(1)
if args.dump_config:
print(cfg.to_json()); return
# ββ Run episode βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
if args.mode == "interactive":
history = run_interactive(cfg, seed=args.seed)
else:
if args.actions_file:
with open(args.actions_file) as f:
actions = [int(l.strip()) for l in f if l.strip()]
elif args.actions:
actions = [int(x.strip()) for x in args.actions.split(",") if x.strip()]
else:
print("β Batch mode requires --actions or --actions-file"); sys.exit(1)
history = run_batch(actions, cfg, seed=args.seed)
if not history:
print("No history to evaluate."); return
evaluate_and_print(history, cfg, api_key=args.api_key)
if __name__ == "__main__":
main()
|