PPDRL: A Pretraining-and-Policy-Based Deep Reinforcement Learning Approach for QoS-Aware Service Composition

Table 2

Hyperparameters for each algorithm.

Algorithm	Hyperparameters
MCOP_M	N/A
GA	population_size: 64, cross_rate: 0.5, mutation_rate: 0.2
PTR	embedding_hidden_dim: 128, LSTM_hidden_dim: 128, lr1_decay_step: 1000, lr1_decay_rate: 0.96, : 0.9
QLR	learning_rate: 0.2, reward_decay: 0.9, e_greedy: 0.6
DQN	hidden_dim: 30, lr_start: 0.0002, lr_decay_step: 5000, lr_decay_rate: 0.96, max_epsilon: 0.9, min_epsilon: 0.1, epsilon_increment:0.01, memory_capacity: 300
PPDRL	preipotrain_lr: 0.001, rl_lr: 0.0001, hidden_dim: 128