def start_proc(dir_seed):
print(dir_seed[1])
rows = ROWS
cols = COLS
num_plant_types = PlantType().num_plant_types
depth = num_plant_types + 3 # +1 for 'earth' type, +1 for water, +1 for health
sector_rows = SECTOR_ROWS
sector_cols = SECTOR_COLS
prune_window_rows = PRUNE_WINDOW_ROWS
prune_window_cols = PRUNE_WINDOW_COLS
garden_step = STEP
action_low = 0
action_high = 1
obs_low = 0
obs_high = rows * cols
garden_days = 100
sector_obs_per_day = int(NUM_PLANTS +
PERCENT_NON_PLANT_CENTERS * NUM_PLANTS)
collection_time_steps = sector_obs_per_day * garden_days # 210 sectors observed/garden_day * 200 garden_days
water_threshold = 1.0
data_collection = DataCollection()
data_collection.evaluate_policy(
data_collection.init_env(rows, cols, depth, sector_rows, sector_cols,
prune_window_rows, prune_window_cols,
action_low, action_high, obs_low, obs_high,
collection_time_steps, garden_step,
num_plant_types, dir_seed[0], dir_seed[1]),
analytic_policy.policy, collection_time_steps, sector_rows,
sector_cols, prune_window_rows, prune_window_cols, garden_step,
water_threshold, sector_obs_per_day)
def __init__(self,
max_time_steps,
rows,
cols,
sector_rows,
sector_cols,
prune_window_rows,
prune_window_cols,
seed=None,
step=1,
dir_path="/"):
super(SimAlphaGardenWrapper, self).__init__(max_time_steps)
self.rows = rows
self.cols = cols
self.num_sectors = (rows * cols) / (sector_rows * sector_cols)
self.sector_rows = sector_rows
self.sector_cols = sector_cols
self.prune_window_rows = prune_window_rows
self.prune_window_cols = prune_window_cols
self.step = step
self.PlantType = PlantType()
self.seed = seed
self.reset()
self.curr_action = -1
self.config = configparser.ConfigParser()
self.config.read('gym_config/config.ini')
# Amount to water every square in a sector by
self.irr_actions = {
1: MAX_WATER_LEVEL,
}
self.plant_centers_original = []
self.plant_centers = []
self.non_plant_centers_original = []
self.non_plant_centers = []
self.centers_to_execute = []
self.actions_to_execute = []
self.plant_radii = []
self.plant_heights = []
self.dir_path = dir_path
def __init__(self,
max_time_steps,
rows,
cols,
sector_rows,
sector_cols,
prune_window_rows,
prune_window_cols,
seed=None,
step=1,
dir_path="",
randomize_seed_coords=False,
plant_seed_config_file_path=None):
"""AlphaGarden's wrapper for Gym, inheriting basic functions from the WrapperEnv.
Args:
max_time_steps (int): the number of time steps a simulator runs before resetting.
rows (int): Amount rows for the grid modeling the garden (N in paper).
cols (int): Amount columns for the grid modeling the garden (M in paper).
sector_rows (int): Row size of a sector.
sector_cols (int): Column size of a sector.
prune_window_rows (int): Row size of pruning window.
prune_window_cols (int): Column size of pruning window
seed (int): Value for "seeding" numpy's random state generator. NOT SEED OF PLANT.
step (int): Distance between adjacent points in grid.
dir_path (str): Directory location for saving experiment data.
randomize_seed_coords (bool): Flag to randomize seed coordinates for given plant seed config file.
plant_seed_config_file_path (str): File path for plant seed configuration.
"""
self.max_time_steps = max_time_steps
super(SimAlphaGardenWrapper, self).__init__(max_time_steps)
self.rows = rows
self.cols = cols
#: int: Number of sectors (representing the area observable to the agent at time t) in garden.
self.num_sectors = (rows * cols) / (sector_rows * sector_cols)
self.sector_rows = sector_rows
self.sector_cols = sector_cols
self.prune_window_rows = prune_window_rows
self.prune_window_cols = prune_window_cols
self.step = step
self.randomize_seed_coords = randomize_seed_coords
self.plant_seed_config_file_path = plant_seed_config_file_path
self.PlantType = PlantType(
) #: :obj:`PlantType`: Available types of Plant objects (modeled).
self.seed = seed
self.reset() #: Reset simulator.
self.curr_action = -1 #: int: Current action selected. 0 = no action, 1 = irrigation, 2 = pruning
#: Configuration file parser for reinforcement learning with gym.
self.config = configparser.ConfigParser()
self.config.read('gym_config/config.ini')
#: dict of [int,str]: Amount to water every square in a sector by.
self.irr_actions = {
1: MAX_WATER_LEVEL,
}
self.plant_centers_original = [
] #: Array of [int,int]: Initial seed locations [row, col].
self.plant_centers = [
] #: Array of [int,int]: Seed locations [row, col] for sectors.
self.non_plant_centers_original = [
] #: Array of [int,int]: Initial locations without seeds [row, col].
self.non_plant_centers = [
] #: Array of [int,int]: Locations without seeds [row, col] for sectors.
self.centers_to_execute = [
] #: Array of [int,int]: Locations [row, col] where to perform actions.
self.actions_to_execute = [] #: List of int: Actions to perform.
self.plant_radii = [
] #: List of tuples (str, float): Tuple containing plant type it's plant radius.
self.plant_heights = [
] #: List of tuples (str, float): Tuple containing plant type it's plant height.
self.dir_path = dir_path
plt.legend()
plt.savefig(save_dir + 'water_use_' + str(trial) + '.png',
bbox_inches='tight',
pad_inches=0.02)
plt.close()
if __name__ == '__main__':
import os
cpu_cores = [i for i in range(0, 80)] # Cores (numbered 0-11)
os.system("taskset -pc {} {}".format(",".join(str(i) for i in cpu_cores),
os.getpid()))
rows = 150
cols = 300
num_plant_types = PlantType().num_plant_types
depth = num_plant_types + 3 # +1 for 'earth' type, +1 for water, +1 for health
sector_rows = 15
sector_cols = 30
prune_window_rows = 5
prune_window_cols = 5
garden_step = 1
action_low = 0
action_high = 1
obs_low = 0
obs_high = rows * cols
garden_days = args.days
sector_obs_per_day = int(NUM_PLANTS +
PERCENT_NON_PLANT_CENTERS * NUM_PLANTS)
# INPUT {type: {(x, y), radius}, ..., {}}
real_data = {
'borage': {
((2, 2), 10),
((5, 5), 10),
},
'sorrel': {
((8, 8), 10),
((11, 11), 10),
},
}
timestep = 10
plant_type = PlantType()
plant_types = plant_type.plant_names
plant_objs = plant_type.get_plant_seeds(0,
ROWS,
COLS,
SECTOR_ROWS,
SECTOR_COLS,
start_from_germination=False,
existing_data=real_data,
timestep=timestep)
plants = [{} for _ in range(len(plant_types))]
grid = np.empty((ROWS, COLS),
dtype=[('water', 'f'), ('health', 'i'), ('nearby', 'O')])
grid['water'] = np.random.normal(0.4, 0.1, grid['water'].shape)