def test_no_5_some_batch_data(self):
agents = get_SmartAgents()
actual_phase = 0
orange_phase_duration = 2
phase_duration = orange_phase_duration + 1
den = [0, 0, 10, 8]
den_pred = [2, 2, 2, 78]
state_to_predict = LearningState(
actual_phase=actual_phase,
starting_actual_phase=actual_phase,
phase_duration=phase_duration,
global_densities=den_pred + [0, 0],
densities=den,
orange_phase_duration=orange_phase_duration)
for i in range(150):
# utrzymanie fazy - gorszy wybór
action_0 = [0]
env = single_simulate(agents,
actual_phase,
phase_duration,
den,
orange_phase_duration=orange_phase_duration,
actions=action_0,
u=env_settings.u_all_2)
for x in range(40):
simulate_from_env(env, action_0)
action_1 = [1]
# zmiana fazy - lepszy wybór
# print('stan bazowy', env.x[env.t])
# print(env.global_memories[-1])
simulate_from_env(env, action_1)
# print('stan', env.x[env.t])
# print('stan po', env.x[env.t])
simulate_from_env(env, [orange])
# print('stan', env.x[env.t])
simulate_from_env(env, [orange])
# print('stan', env.x[env.t])
for x in range(20):
simulate_from_env(env, action_1)
# print('stan', env.x[env.t])
memories = env.global_memories
memories += env.global_memories
save_batches(agents)
batches = get_batches()
train(agents=agents)
env.global_memories = []
for agent in env.agents:
agent.memories = []
prediction = agents[0].model.predict(
state_to_predict.to_learn_array())
# zbiega to 10,20, cokolwiek
print('prediction :)', prediction)
exportData = ExportData(learningMethod='DQN',
learningEpochs=0,
nets=memories,
netName='net14',
densityName='test_fixed_no_4')
exportData.saveToJson()
a = 23
pred_history = []
y_batch_history = []
Globals().u_value = 5
runs = range(len(Globals().val_params))
for run in runs:
Globals().pred_plot_memory = []
Globals().run_no = run
results = []
timeToLearn = 500000
startTime = timer()
generate_my_epochs(learntAgents=False,
epochs=range(Globals().vp().first_epochs_range)
) # bierze nowych agentow i tu jest 'is'
train(learntAgents=False, max_time_learn=Globals().vp().max_time_learn)
# run_learnt_greedy()
lurns = 0
eps_decay = 0
actual_number = 0
while timer() - startTime < timeToLearn:
actual_number += 1
eps_decay += 0.07
Globals().epsilon = 1 - eps_decay
if Globals().epsilon < 0.2:
Globals().epsilon = 0.2
# print('epsilon', Globals().epsilon)
# print('czas', timer() - startTime)
# print('U', Globals().u_value)
generate_my_epochs(learntAgents=True,
epochs=range(2),
from timeit import default_timer as timer
from nn_trainer import train
from random_epochs_generator import generate_random_epochs
from services.globals import Globals
from services.runnerService import run_learnt_greedy
Globals().pred_plot_memory = []
results = []
timeToLearn = 60
startTime = timer()
generate_random_epochs(learntAgents=False, epochs=range(50))
train(learntAgents=False)
run_learnt_greedy()
while timer() - startTime < timeToLearn:
print('czas', timer() - startTime)
generate_random_epochs(learntAgents=True, epochs=range(50))
train(max_time_learn=30)
result = run_learnt_greedy()
results.append(result)
def test_no_5_some_batch_data(self):
agents = get_SmartAgents()
actual_phase = 0
orange_phase_duration = 2
phase_duration = orange_phase_duration + 1
predictions_hist = []
# den = [0, 0, 0, 0]
# den_pred = [2, 2, 2, 78]
pred_x = np.array([[2, 2, 2, 78, 0]])
# state_to_predict = LearningState(actual_phase=actual_phase, starting_actual_phase=actual_phase,
# phase_duration=phase_duration, global_densities=den_pred + [0, 0],
# densities=den,
# orange_phase_duration=orange_phase_duration)
a = 0
for i in range(5000):
a += 1
# utrzymanie fazy - gorszy wybór
action_0 = [0]
# env = single_simulate(agents, actual_phase, phase_duration, den,
# orange_phase_duration=orange_phase_duration,
# actions=action_0, u=env_settings.u_all_2)
env = Env(agents)
for x in range(40):
simulate_from_env(env, action_0)
action_1 = [1] # 41
# zmiana fazy - lepszy wybór
# print('stan bazowy', env.x[env.t])
# print(env.global_memories[-1])
if a % 2 == 0:
action = [1]
else:
action = [0]
simulate_from_env(env, action) # 41
# print('stan', env.x[env.t])
# print('stan po', env.x[env.t])
simulate_from_env(env, [orange]) # 42
# print('stan', env.x[env.t])
simulate_from_env(env, [orange]) # 43
# print('stan', env.x[env.t])
for x in range(20):
simulate_from_env(env, action)
# print('stan', env.x[env.t])
memories = env.global_memories
# print(f'rewardy t=40:{env.agents[0].memories[40].reward}')
# print(f'rewardy t=41:{env.agents[0].memories[41].reward}')
# print(f'rewardy t=42:{env.agents[0].memories[42].reward}')
# memories += env.global_memories
save_batches(agents)
batches = get_batches()
print('x batch', batches[0]['y_batch'][40])
train(agents=agents)
env.global_memories = []
for agent in env.agents:
agent.memories = []
prediction = agents[0].model.predict(pred_x)
predictions_hist.append(prediction)
# zbiega to 10,20, cokolwiek
print('prediction :)', prediction)
exportData = ExportData(learningMethod='DQN',
learningEpochs=0,
nets=memories,
netName='net14',
densityName='test_fixed_no_4')
# exportData.saveToJson()
a = 23
# plt.plot([pred[0][0] for pred in Globals().pred_plot_memory], color='red', label='0')
# plt.plot([pred[0][1] for pred in Globals().pred_plot_memory], color='green', label='1')
plt.plot([x[0][0] for x in predictions_hist],
color='red',
label='0')
plt.plot([x[0][1] for x in predictions_hist],
color='green',
label='1')
plt.legend()
plt.title('Cokolwiek')
plt.savefig('pred.png')
plt.close()
from runner_learnt import run_learnt_greedy
import matplotlib.pyplot as plt
from services.globals import Globals
runs = [0, 1, 2]
for run in runs:
Globals().pred_plot_memory = []
Globals().run_no = run
results = []
timeToLearn = 600
startTime = timer()
generate_random_epochs(
learntAgents=False,
epochs=range(50)) # bierze nowych agentow i tu jest 'is'
train()
run_learnt_greedy()
while timer() - startTime < timeToLearn:
print('czas', timer() - startTime)
generate_random_epochs(learntAgents=True, epochs=range(50))
train(max_time_learn=40)
result = run_learnt_greedy()
results.append(result)
# print(times)
plt.plot([res[0] for res in results])
plt.title('Średnia wszystkich nagród - akcje wedle wyuczonej strategii')
plt.savefig('rewards_mean' + str(run) + '.png')
plt.close()
# plt.show() # rewards_mean
plt.plot([res[1] for res in results])
plt.title('Suma nagród - akcje wedle wyuczonej strategii')
from timeit import default_timer as timer
from nn_trainer import train
from services.drawerService import draw_rewards_mean, draw_rewards, draw_cars_out
from services.globals import Globals
from services.runnerService import generate_random_epochs, run_learnt_greedy
results = []
timeToLearn = 500000
startTime = timer()
agents = generate_random_epochs(learntAgents=False,
epochs=range(Globals().vp.first_epochs_range)
) # bierze nowych agentow i tu jest 'is'
train(learntAgents=False, max_time_learn=Globals().vp.max_time_learn)
result = run_learnt_greedy()
eps_decay = 0
max_iterations_without_progress = 15
iterations_without_progress = 0
while timer() - startTime < timeToLearn:
eps_decay += 0.04
Globals().epsilon = 1 - eps_decay
if Globals().epsilon < 0.2:
Globals().epsilon = 0.2
print(
f'Czas:{round(timer() - startTime, 0)} Epsilon:{round(Globals().epsilon, 2)} Średnia liczba wjeżdżających pojazdów:'
f':{round(Globals().u_value, 2)}')
generate_random_epochs(learntAgents=True,
epochs=range(Globals().vp.epochs_range))
train(max_time_learn=Globals().vp.max_time_learn)
result = run_learnt_greedy()
maximum_possible_cars_out = result[4]
if result[2] > maximum_possible_cars_out * 0.93: # cars_out
# train(learntAgents=False, max_time_learn=Globals().vp().max_time_learn)
# result=run_learnt_greedy()
lurns = 0
eps_decay = 0
while timer() - startTime < timeToLearn:
eps_decay += 0.04
Globals().epsilon = 1 - eps_decay
if Globals().epsilon < 0.2:
Globals().epsilon = 0.2
print('epsilon', Globals().epsilon)
print('czas', timer() - startTime)
print('U', Globals().u_value)
generate_random_epochs(learntAgents=True,
epochs=range(Globals().vp().epochs_range))
draw_batches('static_files/x_batch' + str(Globals().greedy_run_no))
train(max_time_learn=Globals().vp().max_time_learn)
result = run_learnt_greedy()
draw_batches_from_agent(result[3],
file_name='static_files/x_batch_greedy' +
str(Globals().greedy_run_no))
maximum_possible_cars_out = Globals().u_value * Globals().vp(
).max_time_greedy * 8
print('max possible', maximum_possible_cars_out)
if result[2] > maximum_possible_cars_out * 0.93: # cars_out
print('u przed', Globals().u_value)
Globals().u_value = Globals().u_value * 1.2
# print('u po',Globals().u_value)
results.append(result)
lurns += 1
name = 'teraz' + str(Globals().greedy_run_no) + "time" + str(
timer() - startTime) + " " + str(Globals().vp())