LunarLander-v2-run_batch

LunarLander-v2-run_batch.py

def run_batch(self): # train every self.Training_batch episodes
    state = self.env.reset()
    state = np.reshape(state, [1, self.state_size[0]])
    done, score, SAVING = False, 0, ''
    while True:
        # Instantiate or reset games memory
        states, next_states, actions, rewards, predictions, dones = [], [], [], [], [], []
        for t in range(self.Training_batch):
            self.env.render()
            # Actor picks an action
            action, action_onehot, prediction = self.act(state)
            # Retrieve new state, reward, and whether the state is terminal
            next_state, reward, done, _ = self.env.step(action)
            # Memorize (state, action, reward) for training
            states.append(state)
            next_states.append(np.reshape(next_state, [1, self.state_size[0]]))
            actions.append(action_onehot)
            rewards.append(reward)
            dones.append(done)
            predictions.append(prediction)
            # Update current state
            state = np.reshape(next_state, [1, self.state_size[0]])
            score += reward
            if done:
                self.episode += 1
                average, SAVING = self.PlotModel(score, self.episode)
                print("episode: {}/{}, score: {}, average: {:.2f} {}".format(self.episode, self.EPISODES, score, average, SAVING))

                state, done, score, SAVING = self.env.reset(), False, 0, ''
                state = np.reshape(state, [1, self.state_size[0]])

        self.replay(states, actions, rewards, predictions, dones, next_states)