states = np.reshape(states, [1, state_size])
action = QN.act(states)
next_state, rewards, overall_err = sim_env.Assign_Cores(action)
next_state = np.reshape(next_state, [1, state_size])
QN.remember(states, action, rewards, next_state)
states = next_state
if len(QN.memory) > batch_size:
QN.replay(batch_size)
loss_overall = np.append(loss_overall, QN.loss_avg/training)
QN.loss_avg = 0
sim_env.reset()
for u in range(testing):
states = np.reshape(states, [1, state_size])
action = QN.act_test(states)
# print('SNR:', sim_env.SNR[-1])
# print('action:', sim_env.action[action])
next_state, rewards, overall_err = sim_env.Assign_Cores(action)
error = np.append(error, overall_err)
next_state = np.reshape(next_state, [1, state_size])
states = next_state
print(e)
print(sim_env.error/testing)
error_avg = np.append(error_avg, np.power(10, -sim_env.error/testing))
# Save Error and Losses in CSV file, Save weights of networks ####
parameters = '_DQN_S{}_rho{}_SNR{}_PS{}_lr{}_df{}_sl{}_nhl{}_ef{}'.\
format(sim_env.S, sim_env.p, sim_env.SNR_avg[0], sim_env.pi, QN.learning_rate,
QN.gamma, QN.size_layers, QN.number_hidden_layers, QN.epsilon_decay)
print(parameters)
print(r)
parameters = 'old/DQN__S{}_rho{}_SNR{}_PS{}_W4_lr0.0001_df0.0_sl24_nhl1_ef0.9'.format(
r, 0.9, SNR, sim_env.pi)
QN.load(sim_env.channel_type + '/' + parameters)
sim_env = Simulation(number_of_servers=num_of_servers,
number_of_users=1,
historic_time=hist_timeslots,
snr_set=avg_SNR,
csi=0,
channel=0.9)
sim_env.reset()
states = sim_env.state # get first state
for u in range(testing):
states = np.reshape(
states, [1, state_size]) # reshape state to vector for network
action = QN.act_test(states) # get action from DQN agent
# print('SNR:', sim_env.SNR[-1])
# print('action:', sim_env.action[action])
next_state, rewards, overall_err = sim_env.Assign_Cores(
action, u) # get next state, reward and error
error = np.append(error, overall_err)
next_state = np.reshape(next_state,
[1, state_size]) # reshape next state
states = next_state # state = next state
print(r)
print(np.power(10, -sim_env.error / testing))
error_avg = np.append(error_avg, np.power(10, -sim_env.error / testing))
# Save Error and Losses in CSV file, Save weights of networks ####
parameters = 'DQN_S{}_rho{}_SNR{}_PS{}_OverMaxBlkL'.format(
sim_env.S, sim_env.p, sim_env.SNR_avg[0], sim_env.pi)
