INNER CODE UNIT · Python

choose_action

rasbt/machine-learning-book · ch19/cartpole/main.py:68

    def choose_action(self, state):
        if np.random.rand() <= self.epsilon:
            return np.random.choice(self.action_size)
        with torch.no_grad():
            q_values = self.model(torch.tensor(state, dtype=torch.float32))[0]
        return torch.argmax(q_values).item()  # returns action

    def _learn(self, batch_samples):
        batch_states, batch_targets = [], []
        for transition in batch_samples:
            s, a, r, next_s, done = transition

            with torch.no_grad():
                if done:
                    target = r
                else:
                    pred = self.model(torch.tensor(next_s, dtype=torch.float32))[0]
                    target = r + self.gamma * pred.max()

View source record →