routing_modes = [ "Q_routing_2_hop", "Q_routing_1_hop", "Q_routing_0_hop", "Q_routing", "TTSPWRR", "TTSP" ] network_names = ["5x6", "UES_Manhatan", "toronto"] gpu_num = int(sys.argv[1]) algorithm_num = int(sys.argv[2]) network_num = int(sys.argv[3]) config = Config() config.use_GPU = True assert (torch.cuda.is_available()) config.device = torch.device(gpu_num) config.routing_mode = routing_modes[algorithm_num] network_name = network_names[network_num] config.training_mode = True config.does_need_network_state = config.routing_mode in [ "Q_routing_2_hop", "Q_routing_1_hop", "Q_routing_0_hop" ] config.does_need_network_state_embeding = config.routing_mode in [ "Q_routing_2_hop", "Q_routing_1_hop" ] config.retain_graph = config.does_need_network_state_embeding # config.exp_name=config.routing_mode config.should_load_model= False if config.routing_mode== "TTSPWRR" or \
config = Config() config.seed = 1 config.num_episodes_to_run = 8000 # config.file_to_save_data_results = "results/data_and_graphs/VEC.pkl" # config.file_to_save_results_graph = "results/data_and_graphs/VEC.png" config.show_solution_score = False config.visualise_individual_results = False config.visualise_overall_agent_results = False config.standard_deviation_results = 1.0 config.runs_per_agent = 1 config.use_GPU = True config.overwrite_existing_results_file = False config.randomise_random_seed = True config.save_model = False config.device = "cuda:0" config.hyperparameters = { "DQN_Agents": { "learning_rate": 0.00002, "batch_size": 256, "buffer_size": 100000, "epsilon_decay_rate_denominator": 150, "discount_rate": 0.99, "incremental_td_error": 1e-8, "update_every_n_steps": 1, "linear_hidden_units": [1000,800], "final_layer_activation": None, "batch_norm": False, "gradient_clipping_norm": 5, "HER_sample_proportion": 0.8,