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()
#
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)
bg.play_4dqn(episode=10000)