crop-coach 2.0.4

crop_coach : A crop growth simulation model turn into OpenAI Gym environment

CropCoach

Developed by Noureddine Ech-chouky, AgriEdge(c) 2022

Initialing CropCoach with default parameters:

import gym



# -- Initializing the environment : with the default parameters

env = gym.make("CropCoach-v0")

Examples of How To Use (Alpha Version)

Initiation the environment

import gym



# -- Initializing the environment : with the default parameters

env = gym.make("CropCoach-v0")



# -- Sample from observation space :

print(env.observation_space.sample())



# -- Sample from action space :

print(env.action_space.sample())

Simple Rl loop

import gym

import CropCoach



import shutup



shutup.please()





# -- Define the env :

env = gym.make("CropCoach-v0")







# -- Test the env : the rewarding mecanism

episodes = 10

for episode in range(1, episodes + 1):

    state = env.reset()

    done = False

    score = 0



    while not done:

        # env.render()

        action = env.action_space.sample()

        # action = np.array([1000])

        print(f"----------- action : {action} -------------\n")



        n_state, reward, done, info = env.step(action)

        print(f"----------- done : {done} -------------\n")

        print(f"----------- state : {n_state} -------------\n")

        print(f"----------- reward : {reward} -------------\n")

        score += reward



    print(f"Episode --:{episode} Score --:{score}")

Train agents Using stable baselines3 : you need to install stable baselines3 and configure wandb platform, see : my wofost project

# -- Importing dependencies :

import gym

from CropCoach

import numpy as np

import random

import os

import time

from pprint import pprint

from stable_baselines3.common.callbacks import BaseCallback

from stable_baselines3 import A2C, PPO





def create_list_dir(paths_list):

    for path in paths_list:

        if not os.path.exists(path):

            os.makedirs(path)





# logs folder :

from datetime import datetime



# datetime object containing current date and time

now = datetime.now().strftime("%a_%b_%Y_%I_%M_%p")





Test_Objective = "Default_PPO_vs_A2C_MultiDiscrete"



logs_dir = f"logs_{Test_Objective}_{now}"

models_dir = f"models_{Test_Objective}_{now}"

paths_list = [logs_dir, models_dir]



# create the logs and models folders :

create_list_dir(paths_list)





# # -- Init the env  :

env = gym.make("wofost-v0")



# -- Use weights and biases track training and evaluation :

import wandb

from wandb.integration.sb3 import WandbCallback





# -- Init wandb configuration :

config = {"policy_type": "MlpPolicy", "total_timesteps": 10000}



experiment_name = f"{Test_Objective}_{now}"





run = wandb.init(

    name=experiment_name,

    project="Paper_Experiments",

    config=config,

    sync_tensorboard=True,  # auto-upload sb3's tensorboard metrics

    save_code=True,  # optional

)





TIMESTEPS = config["total_timesteps"]



a2c_agent = A2C("MlpPolicy", env, tensorboard_log=logs_dir, verbose=1)



ppo_agent = PPO("MlpPolicy", env, tensorboard_log=logs_dir, verbose=1)





print("Training the A2C agent...")

a2c_agent.learn(

    total_timesteps=TIMESTEPS,

    tb_log_name="A2C",

    callback=WandbCallback(

        gradient_save_freq=2,

        log="all",

        verbose=1,

    ),

)

a2c_agent.save(f"{models_dir}/a2c_{TIMESTEPS}")

print("Done training the A2C agent")



print("Training the PPO agent...")

ppo_agent.learn(

    total_timesteps=TIMESTEPS,

    tb_log_name="PPO",

    callback=WandbCallback(

        gradient_save_freq=2,

        log="all",

        verbose=1,

    ),

)

ppo_agent.save(f"{models_dir}/ppo_{TIMESTEPS}")

print("Done training the PPO agent")



run.finish()



env.close()

Changing the default parameters :

import gym

import crop_coach



import shutup



shutup.please()











crop_path = "C:/Users/noureddine/Desktop/default_data/crop/crop.cab"

site_path = "C:/Users/noureddine/Desktop/default_data/site/site.cab"

soil_path = "C:/Users/noureddine/Desktop/default_data/soil/soil.cab"







args={

        "files_paths": {

            "site":site_path, "soil":soil_path, "crop":crop_path

            },

        "sample_year":True,

        "Agromanager_dict":{

            "crop_name": "wheat","crop_variety": "Winter_wheat_101","campaign_start_date": "-01-01","crop_start_type":"emergence","emergence_date": "-03-31","crop_end_type": "harvest","harvest_date": "-08-11", "max_duration": 300

        },

        "Costs_dict": {"Irrigation": 150, "N": 8, "P": 8.5, "K": 7, "Selling": 2.5},

        "Discount_factors_dict": {"Irrigation": 1, "N": 1, "P": 1, "K": 1},

        "year": 2019,

        "years_count": 1,

    }



# -- Define the env :

env = gym.make("CropCoach-v0", **args)



# -- Test the env : the rewarding mecanism

episodes = 10

for episode in range(1, episodes + 1):

    state = env.reset()

    done = False

    score = 0



    while not done:

        env.render()

        action = env.action_space.sample()

        # action = np.array([1000])

        print(f"----------- action : {action} -------------\n")



        n_state, reward, done, info = env.step(action)

        print(f"----------- done : {done} -------------\n")

        print(f"----------- state : {n_state} -------------\n")

        print(f"----------- reward : {reward} -------------\n")

        score += reward



    print(f"Episode --:{episode} Score --:{score}")