Table 1 Parameters of the computational model
From: Leveraging agent-based models and deep reinforcement learning to predict taxis in cell migration
Hyperparameter | Value/Description |
|---|---|
Batch size | 512 (number of transitions sampled from the replay buffer) |
Discount factor (γ) | 0.1 (importance of future rewards) |
Epsilon (ϵ) | 0.9, decays to 0.0001 over 1000 episodes |
Learning rate | 1 × 10−4 (for the AdamW optimizer) |
Soft update rate (Ï„) | 0.005 (rate at which the target network is updated) |
Number of observations (No) | 72 (NN input) |
Number of actions (Na) | 16 (NN output) |
Time step (Δt) | 10 min |
Cell velocity (vc) | 0.2 μm/min (mean velocity of the cell from17) |
Radius of the cell (Rc) | 10 μm |
Radius of the cell nucleus (RN) | 4 μm |
