def __init__(self, agent_ids):
self.agent_ids = agent_ids
self.flat_env = BombermanEnv(agent_ids)
self.observation_space = spaces.Tuple((spaces.Box(low=0, high=1, shape=(15, 15, 14)),
spaces.Box(low=0, high=1, shape=(4,)),
#spaces.MultiBinary(3),
#spaces.Box(low=0, high=1, shape=(1,)),
spaces.MultiBinary(len(HIGH_LEVEL_ACTIONS))))
self.action_space = spaces.Discrete(3)
self.high_level_mode = True
self.action_buffer = {}
self.high_low_mapping = {}
def __init__(self,
orders,
products,
start_date,
end_date,
stock_policy=None,
debug=True,
reward_function=None):
"""
init the env by using env = gym.make("Procurement-v0" ....)
:param orders: Array of orders
:type orders: []Order (see class Order)
:param products: Array of products define the initial stock
:type products: []Product (see class Product)
:param start_date: Start date of the env
:type datetime
:param end_date: End date of the env
:type datetime
:param stock_policy: Custom stock policy
:type dict
:param debug: enable console prints
:type bool
:param reward_function: custom reward function - see example_reward function
:type function
:return: env
"""
if reward_function:
self.reward_function = reward_function
else:
self.reward_function = self.example_reward
self.orders = orders
self.products = products
self.start_date = start_date
self.current_date = start_date
self.end_date = end_date
if stock_policy:
self.stock_policy = stock_policy
self.debug = debug
# Environment OpenAI metadata
self.reward_range = (-np.inf, 0)
self.action_space = spaces.Box(
low=0.0,
high=self.get_max_product_order_amount(),
shape=(len(products), self.get_max_product_range()),
dtype=np.int) # products x order amounts
self.observation_space = spaces.Box(
low=0.0,
high=self.get_max_product_order_amount(),
shape=((end_date - start_date).days, len(products), 1),
# matrix of days, products, amount
dtype=np.int)
stock = {}
for p in products:
stock[p.get_id()] = p.get_initial_stock()
self.stock = stock
if debug:
print("init finished")
import numpy as np
import ray
from gym.vector.utils import spaces
from ray import tune
from ray.rllib.models import ModelCatalog
from ray.rllib.models.torch.fcnet import FullyConnectedNetwork as TorchFC
from ray.rllib.models.torch.torch_modelv2 import TorchModelV2
from ray.rllib.utils.framework import try_import_torch
from ray.tune import Callback
from environment.cartpole_ray import CartPoleEnv
torch, nn = try_import_torch()
custom_input_space = spaces.Box(low=-np.inf,
high=np.inf,
shape=(2, ),
dtype=np.float32)
class TorchCustomModel(TorchModelV2, nn.Module):
"""Example of a PyTorch custom model that just delegates to a fc-net."""
def __init__(self, obs_space, action_space, num_outputs, model_config,
name):
TorchModelV2.__init__(self, custom_input_space, action_space,
num_outputs, model_config, name)
nn.Module.__init__(self)
self.torch_sub_model = TorchFC(custom_input_space, action_space,
num_outputs, model_config, name)
def forward(self, input_dict, state, seq_lens):
from typing import Tuple
from gym.vector.utils import spaces
from ray.rllib import MultiAgentEnv
from ray.rllib.utils.typing import MultiAgentDict
from training.hierarchical_learning.bomberman_multi_env import *
import numpy as np
COLLECT_OBSERVATION_SPACE = spaces.Tuple((spaces.Box(low=0, high=1, shape=(15, 15, 11)), spaces.MultiBinary(6)))
DESTROY_OBSERVATION_SPACE = spaces.Tuple((spaces.Box(low=0, high=1, shape=(15, 15, 11)), spaces.MultiBinary(6)))
KILL_OBSERVATION_SPACE = spaces.Tuple((spaces.Box(low=0, high=1, shape=(15, 15, 13)), spaces.MultiBinary(6)))
HIGH_LEVEL_ACTIONS = ['COLLECT', 'DESTROY', 'KILL']
class HierarchicalBombermanMultiEnv(MultiAgentEnv):
def __init__(self, agent_ids):
self.agent_ids = agent_ids
self.flat_env = BombermanEnv(agent_ids)
self.observation_space = spaces.Tuple((spaces.Box(low=0, high=1, shape=(15, 15, 14)),
spaces.Box(low=0, high=1, shape=(4,)),
#spaces.MultiBinary(3),
#spaces.Box(low=0, high=1, shape=(1,)),
spaces.MultiBinary(len(HIGH_LEVEL_ACTIONS))))
self.action_space = spaces.Discrete(3)
self.high_level_mode = True
self.action_buffer = {}
self.high_low_mapping = {}
def reset(self):