from utils.net import simple_net
import pickle
from keras.models import load_model
seed = 1234
np.random.seed(seed)
def demand():
while True:
d = int(np.random.normal(10, 5))
if d < 0:
d = 0
yield d
demand_gen = demand()
#
env_exp3 = BeerGameEnv(demand_gen, lag=2)
api = env_exp3.start_play()
state, r, d = next(api)
shape = np.array(state).flatten().shape
#
# agents_exp3 = [Agent(policy="ar")]
# agents_exp3.append(Agent())
# agents_exp3.append(DQN(state_shape=shape, n_action=25, net=simple_net))
# agents_exp3.append(Agent())
#
# bg_exp3 = chain_wrapper(agents_exp3, env_exp3)
# bg_exp3.play_mixed(episode=1000)
#
# plt.plot(np.array(bg_exp3.agents[0].cum_r) + np.array(bg_exp3.agents[1].cum_r)
# + np.array(bg_exp3.agents[2].cum_r) + np.array(bg_exp3.agents[3].cum_r))
# plt.show()
from Beer_game.beer_game_env import BeerGameEnv
import numpy as np
import time
import tqdm
def demand():
while True:
yield np.random.randint(10)
demand_gen = demand()
env = BeerGameEnv(demand_gen, lag=2)
api = env.start_play()
next(api)
env.on_order
a = np.random.randint(10)
api.send(a)
env.cost
env.week
state = env.reset()
action = np.array([0, 0, 0, 0])
env.stock
env.trans
state, cost, done = env.step(action)
import numpy as np
from Beer_game.wrapper import chain_wrapper
import matplotlib.pyplot as plt
import pickle
eps = 10
batch_size = 32
def demand():
while True:
yield np.random.uniform(10)
demand_gen = demand()
env = BeerGameEnv(demand_gen, lag=2)
api = env.start_play()
agents = []
state, r, d = next(api)
r
d
shape = np.array(state).flatten().shape
shape
state
for i in range(4):
agents.append(DQN(state_shape=shape, n_action=10, net=simple_net))
bg = chain_wrapper(agents, env)
seed = 1234
np.random.seed(seed)
def demand():
while True:
d = int(np.random.normal(10, 5))
if d < 0:
d = 0
yield d
demand_gen = demand()
env_exp4 = BeerGameEnv(demand_gen, lag=2)
agents_exp4 = []
for i in range(4):
agents_exp4.append(
DQN(state_shape=(8, ),
n_action=25,
net=simple_net,
model_path='models/dqn4'))
bg_exp4 = chain_wrapper(agents_exp4, env_exp4)
bg_exp4.play(episode=1000000)
plt.plot(
np.array(bg_exp4.agents[0].cum_r) + np.array(bg_exp4.agents[1].cum_r) +
np.array(bg_exp4.agents[2].cum_r) + np.array(bg_exp4.agents[3].cum_r))
plt.savefig('models/cum_r.png')