Java Deep Learning Projects

Calculating the reward

Now that the agent is provided with some guidance—reinforcement—the next task is to calculate the reward for each action the agent makes. Take a look at this code:

// Compute reward for an action float calcReward(float[][] CurrMap, float[][] NextMap) {        int newGoal = calcGoalPos(NextMap);// first, we calculate goal position for each map        if(newGoal == -1) // if goal position is the initial position (i.e. no move)            return (size * size + 1); // we reward the agent to 4*4+ 1 = 17 (i.e. maximum reward)        return -1f; // else we reward -1.0 for each bad move     }