# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under the BSD-style license found in the # LICENSE file in the root directory of this source tree. """ Baseline Inference Script for Smart Farm Irrigation Environment. This script runs a simple rule-based agent through all 3 tasks and produces reproducible baseline scores. The agent uses smart rules considering BOTH moisture AND temperature: - Hot day (>=38C) + dry soil → water more - Rainy day (>15mm) → don't water - Normal rules based on moisture level Usage: python baseline.py python baseline.py --city Mumbai python baseline.py --city Delhi --task 1 """ import argparse import sys import os sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) from server.my_env_environment import FarmEnvironment from graders import grade_task1, grade_task2, grade_task3 from tasks import get_all_tasks def simple_agent(observation: dict) -> str: """ A smart rule-based agent that decides how much to water based on soil moisture, temperature AND rainfall. Rules: - Heavy rain (>15mm) → no water (already raining!) - Very hot (>=38C) + dry → water high - Very hot (>=38C) + okay → water medium - Dry (<40%) → water high - Slightly dry (<55%) → water medium - Slightly moist (<65%) → water low - Good moisture (>=65%) → no water Args: observation: Current farm state dictionary Returns: water_amount string: "none"/"low"/"medium"/"high" """ moisture = observation.get("soil_moisture", 50) rainfall = observation.get("rainfall", 0) temperature = observation.get("temperature", 30) # Rule 1 — Don't water if already raining heavily if rainfall > 15: return "none" # Rule 2 — Very hot day needs more water if temperature >= 38: if moisture < 45: return "high" elif moisture < 60: return "medium" elif moisture < 70: return "low" else: return "none" # Rule 3 — Warm day normal rules if temperature >= 30: if moisture < 40: return "high" elif moisture < 55: return "medium" elif moisture < 65: return "low" else: return "none" # Rule 4 — Cool day needs less water if moisture < 35: return "medium" elif moisture < 50: return "low" else: return "none" def run_task(env, task_id: int, city: str = "Indore") -> dict: """ Runs the agent through a single task and returns results. """ task_config = { 1: {"max_steps": 7, "name": "Basic Moisture Control"}, 2: {"max_steps": 14, "name": "Weather Adaptive Irrigation"}, 3: {"max_steps": 30, "name": "Full Season Management"}, } config = task_config[task_id] max_steps = config["max_steps"] task_name = config["name"] print(f"\n{'='*50}") print(f"Running Task {task_id}: {task_name}") print(f"City: {city} | Max Steps: {max_steps}") print(f"{'='*50}") # Reset environment obs = env.reset(city=city) obs_dict = { "temperature": obs.temperature, "rainfall": obs.rainfall, "soil_moisture": obs.soil_moisture, "crop_health": obs.crop_health.value, "day_number": obs.day_number, "crop_stage": obs.crop_stage.value, } episode_history = [] for step in range(max_steps): # Agent decides action water_amount = simple_agent(obs_dict) action_map = { "none": 0, "low": 1, "medium": 2, "high": 3 } # Record this step BEFORE taking action episode_history.append({ "day": obs_dict["day_number"], "soil_moisture": obs_dict["soil_moisture"], "crop_health": obs_dict["crop_health"], "water_action": action_map[water_amount], "rainfall": obs_dict["rainfall"], "temperature": obs_dict["temperature"], "crop_stage": obs_dict["crop_stage"], }) print(f"Day {obs_dict['day_number']:2d} | " f"Moisture: {obs_dict['soil_moisture']:5.1f}% | " f"Temp: {obs_dict['temperature']:4.1f}C | " f"Rain: {obs_dict['rainfall']:4.1f}mm | " f"Action: {water_amount}") # Take action from models import FarmAction, WaterAmount action = FarmAction(water_amount=WaterAmount(water_amount)) result = env.step(action) # Check if done if result.done: break # Update observation for next step obs_dict = { "temperature": result.temperature, "rainfall": result.rainfall, "soil_moisture": result.soil_moisture, "crop_health": result.crop_health.value, "day_number": result.day_number, "crop_stage": result.crop_stage.value, } # Grade the episode graders = { 1: grade_task1, 2: grade_task2, 3: grade_task3, } score = graders[task_id](episode_history) print(f"\nTask {task_id} Final Score: {score}") return { "task_id": task_id, "task_name": task_name, "score": score, "steps_completed": len(episode_history), "episode_history": episode_history, } def main(city: str = "Indore", task_id: int = None): """ Main function — runs agent through all 3 tasks. """ print("\n" + "="*50) print(" SMART FARM IRRIGATION — BASELINE SCORES") print("="*50) print(f"City: {city}") print(f"Agent: Smart Rule-Based Agent (moisture + temperature)") env = FarmEnvironment() tasks_to_run = [task_id] if task_id else [1, 2, 3] results = [] for tid in tasks_to_run: result = run_task(env, tid, city=city) results.append(result) print("\n" + "="*50) print(" FINAL BASELINE SCORES") print("="*50) for result in results: print(f"Task {result['task_id']} ({result['task_name']}): " f"{result['score']}") avg = sum(r["score"] for r in results) / len(results) print(f"\nAverage Score: {round(avg, 2)}") print("="*50) return results if __name__ == "__main__": parser = argparse.ArgumentParser( description="Run baseline agent on Smart Farm Environment" ) parser.add_argument( "--city", type=str, default="Indore", help="City name for weather data (default: Indore)" ) parser.add_argument( "--task", type=int, default=None, choices=[1, 2, 3], help="Run specific task only (default: run all tasks)" ) args = parser.parse_args() main(city=args.city, task_id=args.task)