def run_train(Env, curr_RL_Config):
# run the training process
"""
Train the agents
"""
Num_neighbor = Env.n_Neighbor
Num_d2d = Env.n_Veh
Num_CH = Env.n_RB
Num_D2D_feedback = curr_RL_Config.Num_Feedback
# construct a BS agent
BS_Agent = Agent(Num_d2d, Num_CH, Num_neighbor, Num_D2D_feedback, Env,
curr_RL_Config)
Num_Episodes = curr_RL_Config.Num_Episodes
Num_Train_Step = curr_RL_Config.Num_Train_Steps
# get the train loss
[Train_Loss, Reward_Per_Train_Step, Reward_Per_Episode,
Train_Q_mean, Train_Q_max_mean, Orig_Train_Q_mean, Orig_Train_Q_max_mean] \
= BS_Agent.train(Num_Episodes, Num_Train_Step)
return [
Train_Loss, Reward_Per_Train_Step, Reward_Per_Episode, Train_Q_mean,
Train_Q_max_mean, Orig_Train_Q_mean, Orig_Train_Q_max_mean
]
def run_train(Env, curr_RL_Config):
# run the training process
"""
Run the Training Process
"""
# parameters to construct a BS Agent object
Num_neighbor = Env.n_Neighbor
Num_d2d = Env.n_Veh
Num_CH = Env.n_RB
Num_D2D_feedback = curr_RL_Config.Num_Feedback
BS_Agent = Agent(Num_d2d, Num_CH, Num_neighbor, Num_D2D_feedback, Env, curr_RL_Config)
Num_Episodes = curr_RL_Config.Num_Episodes
Num_Train_Step = curr_RL_Config.Num_Train_Steps
# get the train loss
[Train_Loss, Train_Q_mean, Train_Q_max_mean] = BS_Agent.train(Num_Episodes, Num_Train_Step)
return [Train_Loss, Train_Q_mean, Train_Q_max_mean]
def load_trained_model(Env, curr_RL_Config):
# load the trained C-Decision model for testing
"""
Load the trained DNN-RL model
"""
# parameters to construct a BS Agent object
Num_neighbor = Env.n_Neighbor
Num_d2d = Env.n_Veh
Num_D2D_feedback = curr_RL_Config.Num_Feedback
Num_CH = Env.n_RB
# construct a BS agent
BS_Agent = Agent(Num_d2d, Num_CH, Num_neighbor, Num_D2D_feedback, Env,
curr_RL_Config)
# load the Trained model weights
# Training Parameters
BATCH_SIZE = curr_RL_Config.Batch_Size
num_episodes = curr_RL_Config.Num_Episodes
num_train_steps = curr_RL_Config.Num_Train_Steps
GAMMA = curr_RL_Config.Gamma
V2I_Weight = curr_RL_Config.v2i_weight
# load the trained results according to their corresponding simulation parameter settings
curr_sim_set = 'Train-Result' + '-Feedback-' + str(Num_D2D_feedback) + '-BatchSize-' + str(BATCH_SIZE) \
+ '-Gamma-' + str(GAMMA) + '-V2Iweight-' + str(V2I_Weight)
folder = os.getcwd() + '\\' + curr_sim_set + '\\'
if not os.path.exists(folder):
os.makedirs(folder)
print('Create the new folder in Testing main ', folder)
model_dir = folder
model_name = 'Q-Network_model_weights' + '-Episode-' + str(num_episodes) \
+ '-Step-' + str(num_train_steps) + '-Batch-' + str(BATCH_SIZE) + '.h5'
model_para = model_dir + model_name
# save the Target Network's weights in case we need it
target_model_name = 'Target-Network_model_weights' + '-Episode-' + str(num_episodes) + '-Step-' \
+ str(num_train_steps) + '-Batch-' + str(BATCH_SIZE) + '.h5'
target_model_para = model_dir + target_model_name
# load Q-Function Network weights
BS_Agent.brain.model.load_weights(model_para)
# load Target Network weights
BS_Agent.brain.target_model.load_weights(target_model_para)
# for debugging
print('Load the trained model successfully under this setting!')
# return the agent with trained model
return BS_Agent